# REPLAY OPTICA ONLINE INDUSTRY MEETING OIM 3D Sensing JUNE 23rd 2026

https://www.youtube.com/watch?v=EzXbbvAzank

[01:05] Hey, hey, hey.
[01:32] Out in the field where the morning breaks,
[01:36] The tractor rolls and the dust it makes.
[01:40] 16 eyes on a wide open view.
[01:48] John Deere driving the whole day through.
[01:51] No hands on the wheel, just lighting time.
[01:54] Finding its way down the orchard line.
[02:03] From the dirt to the chip to the open
[02:07] Road. Photons carry the heavy load.
[02:12] This is the way machines see.
[02:16] Beams of light where the future's going to be.
[02:20] No moving parts, just a steady glow.
[02:23] Lighting up the dark so the world can know.
[02:28] Time a flight on the country breeze. Oh, this is the way machine see.
[02:44] The motive steers without a sound.
[02:48] A meta surface turning light around.
[02:52] Chillers hums with a narrow tone.
[03:00] A tunable laser all its own.
[03:03] Point cloud counts every grain of sand.
[03:08] 4D pictures from a chip in your hand.
[03:16] And flex computer draws a dream in code before the silicone ever hits the road.
[03:24] This is the way machines see.
[03:29] Beams of light where the future's going to be.
[03:32] No moving parts, just a steady blow.
[03:35] Lighting up the dark so the world can blow.
[03:39] Time will fly on a country breeze.
[03:44] Oh, this is the way machines see.
[03:56] Smaller, lighter, cheaper, clear.
[04:00] The road ahead is getting near.
[04:05] Cars and robots, fields and skies, all of it
[04:09] Riding on a beam of light.
[04:13] This is the way machines see.
[04:16] Welcome to the meeting.
[04:19] Join a business conversation.
[04:22] Optic calling and the lights are bright.
[04:26] June 23rd is where we're going to go.
[04:29] Photonics lifting us up like a breeze.
[04:34] Oh, this is the way machines see.
[05:01] Welcome to the last Optica online industry meeting before the summer break.
[05:06] Today's theme is 3D sensing.
[05:11] the next 90 minutes.
[05:13] Contribute to our discussion on AIdriven stereo vision, time of flight, structured light, liar, and the emerging field of photonic radar.
[05:22] Learn how photonics is driving 3D sensing towards mass adoption while cutting weight, size, power, and cost.
[05:34] Time now to go live.
[05:37] Hello everyone and welcome.
[05:39] Welcome to this very special online industry meeting.
[05:41] We are talking about 3D sensing and as you heard with the song, this is the way machines see.
[05:49] We are really lucky today because we have an integrator who is here to tell us about what they need and then we have many many companies offering part of the solution and the magic of these meetings is that we can all of us work together to satisfy the demand of the end user.
[06:05] So let me first of all introduce myself and the team.
[06:07] We are Optica.
[06:10] We are Optica corporate engagement.
[06:10] We are an
[06:12] industry network of companies all over the world and our world or life or passion is to connect all of you.
[06:17] So you can do things that you cannot simply do alone.
[06:20] Today we have two teams like we usually do.
[06:25] one the integrator the end user the company that is here with a shopping list looking to fulfill it with the companies offering 3D sensing solutions and then the enablers the integrator is John Deere and the enablers are the lighter company point cloud the lighter company locomotive the semiconductor laser in heterogeneously integrated semiconductor laser chilas laser and the software of flex compute.
[06:54] we are going to do magic today because all of you will have to collaborate to make sure that we all satisfy the needs of John Deere.
[07:00] Today we do this for free for all of you only for one reason, one reason only because our corporate members support us doing that.
[07:08] If you want to support what we're doing here, make sure that you join as a corporate
[07:13] member.
[07:14] If you're not yet a corporate member,
[07:15] all you have to do is go to optica.org/join corporate and then you can join our fantastic industry network.
[07:22] As you heard in the video, this is our last online industry meeting until summer finishes.
[07:28] And then when summer finishes and then we start back, we come back on the 25th of August.
[07:32] Write this one down.
[07:34] Photonics in ofology and vision care.
[07:37] And we're going to have lots of topics every two weeks on different application fields such as extended reality, quantum safe, new space, quantum computing, and of course the biggest of all go package optics.
[07:48] If you want to present, sponsor or be visible at any of these events, please contact us and we'll be do we'll do our best or very best to make sure that your brand, your company and your needs are showcased in this online industry meeting.
[08:02] For those of you who want to meet face to face and I prefer also face to face, all of us are meeting in Malaga on the 24th and 25th of September at the global photonics economic forum.
[08:10] If you have not reserved
[08:15] your seat yet, we have a limited amount of seats at this event.
[08:20] Please register as soon as possible.
[08:23] Do not complain if one week before the event there are no seats because I do what I can.
[08:28] But we are of course overwhelmed by the demand of the industry.
[08:33] Also on the 21st and 22nd of October, we have our PEC photonic enabled cloud computing event.
[08:38] And on 9th of 10 of December in Paris, we are meeting to discuss one of the most important topics in photonics today.
[08:43] In the current geopolitical situation, it's important to understand what photonics does for defense and aerospace.
[08:51] Also, I would like to say one more thing before I leave the floor to Olga.
[08:55] This meeting is live streaming YouTube.
[08:56] So, let me say hello to the YouTubers in the world.
[08:58] Hello YouTubers in the world.
[09:00] If you have any question or comment, just please write in the chat and I will read it in the room.
[09:03] And this of course also valid for the people here with me in the Zoom room.
[09:07] This is an interactive meeting.
[09:09] It will be as good as you want it to be.
[09:11] So write in the chat what you do, what company you come from.
[09:15] Interact to each other in the chat.
[09:15] The most important
[09:16] thing that we want to do in the next 90 minutes is to make sure that all of you find a potential partner, supplier, customer, invent, investor or maybe somebody to hire.
[09:23] And with this I would like to give the floor to the person who put together this fantastic agenda and who's been doing a fantastic job leading optica this optics design optical integration and even optical instrumentation.
[09:36] Dr. Olga the floor and the attention of everyone is yours.
[09:39] Thank you Jose thank you for the yeah that was a very energetic introduction as always and yeah I'm happy to welcome everyone to this meeting as well as our speakers as well.
[09:51] For sure I would like to thank our support team.
[09:53] It's from the opticus Patel and our IT service Nikico Miller.
[09:59] So without them we wouldn't be live today.
[10:01] And so let's start.
[10:05] Uh this meeting is actually a little bit uh of the review of another meeting that happened two years ago and two years ago the situation was totally different because the entire world was very very
[10:16] hyped about lighter.
[10:19] Lighter especially for the automotive applications.
[10:22] And now since 2 years later let's see the situation looks slightly bit different.
[10:26] and we decided to revisit and see what has changed and what does the future of the technology and generally for the of the 3D sensing because it seems that the big promise that was made for uh for two three years ago didn't hold to the reality.
[10:42] and I would like to also if in this respect we're also invited to revisit one of the speakers from two years ago who is the Zach Banifas from John Deere.
[10:52] already back then I told him Zach you were a little bit you had your thoughts on the how realistic all of this is.
[11:00] and I see that two years later it's interesting to see what you think now and what especially you how you feel this field will develop in another two three five years.
[11:11] so I will welcome to the floor our opening speaker Zach Bonifas from John Deere.
[11:19] Uh, thank you, Olga.
[11:22] Um, by the way, I really enjoyed that uh AI generated song.
[11:25] That was a that was a real treat.
[11:27] I think you guys have a hit on your hands.
[11:29] We'll send you the copy.
[11:31] We'll send you the copy for your uh Skype.
[11:35] Not Skype. Spotify.
[11:37] That would be great.
[11:41] All right.
[11:43] Olga, are you seeing my slides?
[11:43] Yes.
[11:45] Okay.
[11:45] Wonderful.
[11:45] Uh I'm going to start by giving maybe a little bit of perspective on how we use 3D sensing at John Deere.
[11:52] Uh and then I'll talk about um kind of the potential that we see or or really what's important to us when it comes to the various 3D uh sensing technologies.
[12:06] So um at Deer, we we've had 3D sensing on our equipment for about 10 years now.
[12:14] Um, and back in in 2010, I embarked on a
[12:19] project to deliver a system that would use uh a stereo camera to automate the transfer of grain from a harvesting machine to a trailer.
[12:28] And I was given two years uh to accomplish that task.
[12:34] And it only took me five.
[12:34] Okay, so that that was a joke, by the way.
[12:37] Uh but since then we've seen an explosion of perception sensing on our equipment and much of that is uh 3D sensing based and we use perception sensing in every step of our growing cycle uh for broadacre farming but I'm really going to focus on just uh the 3D sensing uh applications for this presentation.
[13:05] So I'm going to start with the prepare step and this generally means tillillage.
[13:10] So what we're trying to do here uh is to uh turn crop residue into the ground so that that residue will
[13:19] decompose and become organic matter for the next crop.
[13:23] And we now have an autonomous tractor uh that that does this tillillage step.
[13:29] And the autonomy system includes 16 cameras for obstacle detection.
[13:36] Uh it has four cameras on all four sides of the roof of the tractor.
[13:43] And these four camera uh setups are really you know a multiple view geometry system that goes beyond you know traditional two camera stereo vision.
[13:56] Uh step number two is to plant and the planting depth and the consistency of the planting depth uh can have a big impact on how much food gets grown and harvested at the end of the growing season.
[14:10] So farmers are very interested in knowing that the seeds are going in at the desired depth.
[14:16] So to measure that
[14:20] Depth, we've developed a camera that uses structured light.
[14:21] And uh based on the shape of a laser line that's projected onto the ground perpendicular to the furrow, uh we are able to sense the depth of that furrow.
[14:36] The third step is to nurture the crop.
[14:39] Uh and we do this by fertilizing uh and killing weeds and pests that are going to compete for water, nutrients and sunlight uh and prevent the crop from achieving its full potential.
[14:53] And for that we have uh again a stereo camera that looks out in front of the self-propelled sprayer to sense the crop rows and to guide the sprayer down the row without running over the crop.
[15:08] And then the last step is harvest.
[15:08] And we use stereo cameras in a couple of different ways in the harvest step uh to make the farmer's job easier and and more productive.
[15:18] First, we use stereo
[15:21] cameras to automate the transfer of grain from the combine to the grain cart.
[15:26] And second, we use stereo cameras to look out in front of the combine to see how much crop the machine is about to ingest.
[15:36] Okay?
[15:40] And that enables us to automatically optimize the speed of the combine.
[15:43] So combine, you know, come uh harvest season, it's not uncommon for farmers to harvest 16 to 18 hours a day.
[15:53] And this makes the uh machine much much more productive.
[16:00] So um you know deer doesn't just make equipment for large acre farming.
[16:08] Um and 3D sensing is becoming more integrated across all of our uh market segments.
[16:15] uh and all of our markets have something in common and that is the scarcity of qualified labor or the availability of
[16:23] any labor uh whatsoever uh in these kind of remote areas where a lot of these these tasks are taking place.
[16:33] So all of our customer segments are asking for autonomy uh or at least smarter machines to help them get their job done.
[16:42] Uh this is happening in our turf business.
[16:44] Uh and this is an exciting development uh because we make these machines in pretty high volume uh relative to our big tractors and combines or construction equipment.
[16:58] Uh when it comes to construction, there are at least a thousand different ways I think that 3D sensing can improve our equipment.
[17:08] But I'll give you a couple of of examples of what we've implemented thus far.
[17:11] The first product I will highlight is a system we call smart detect.
[17:14] So smart detect uses a stereo camera to detect obstacles behind a wheel loader.
[17:24] Uh and wheel loaders drive in reverse a lot and uh construction work sites can be rather chaotic.
[17:33] So a system like this can really improve work site safety.
[17:38] Uh the second product I'll share helps our customers that move dirt for a living.
[17:44] So one of the common tools for moving dirt on a work site is called a scraper.
[17:49] And we offer a system that uses a stereo camera to measure how much dirt is being moved in each load and cumulatively cumulatively excuse me uh how much dirt is being moved over the course of a day.
[18:08] And then orchards uh are a market that have been the earliest adopters of autonomy I would say and we have two uh autonomous orchard offerings.
[18:17] The first is a conventional tractor that will tow a blast sprayer
[18:25] and then the second is a purpose-built self-propelled autonomous blast sprayer.
[18:32] Uh and that machine was originally uh developed by a company called Gus that John Deere uh recently acquired.
[18:40] And both of those machines um heavily leverage LAR sensing.
[18:46] So uh when it comes to these various 3D sensing uh technologies,
[18:55] um I'll start with cameras.
[18:58] And there are a couple of things that we really value in an image sensor.
[19:03] Uh we do a lot of complex multiple view geometry to sense depth.
[19:12] So global shutters are uh strongly preferred to rolling shutter uh for that reason.
[19:16] And uh at the same time we work in very challenging high dynamic range uh conditions and global shutter and HDR is
[19:26] a hard combination to find.
[19:29] Uh but that is something we are definitely interested in.
[19:34] Uh when it comes to lidar,
[19:36] uh my criticism over the years has been that uh it's always been rather expensive and quite poor at penetrating the thick dust that we tend to see in our environment.
[19:49] So that really really limits uh has limited the uh utility of LAR.
[19:56] Now that said, um I think the emergence of LAR at longer wavelengths uh can improve uh dust penetration.
[20:07] So that's helpful.
[20:09] Um and some but not all of our applications uh could benefit from LAR with extreme uh depth accuracy.
[20:18] That's that's hard to find.
[20:22] We're also very excited about the the emergence of FMCW LAR to help classify
[20:30] obstacles as either static or dynamic.
[20:33] Uh that's very helpful for us.
[20:36] And we also see FMCW LAR as potentially an excellent uh source of odometry information.
[20:44] Um, our environment, uh, as I've said in the past, is extremely harsh from a shock and vibration standpoint.
[20:50] So, LAR needs to be designed, uh, with our environmental requirements in mind.
[20:55] And when it comes to cost, um, I'm really encouraged by what I've seen in the past two to three years.
[21:04] And the cheaper that LAR becomes, uh, the more we're going to use it.
[21:12] And I know this is an optica meeting, uh, but I would be remiss if I didn't, uh, talk about radar.
[21:17] Uh, it's really the only sensing technology that we can rely on to penetrate thick dust.
[21:24] Uh, but then we have to live with the uh,
[21:33] resolution of of radar, which is of
[21:36] course going to be orders of magnitude
[21:38] lower than LAR.
[21:40] And also most radars that are out there
[21:43] are really designed for automotive use
[21:45] cases. Um those radars do not work well
[21:49] in our environment. Uh and we have lots
[21:52] of use cases that could benefit from
[21:53] radar. So the radar tech needs to be uh
[21:57] adaptable to a lot of different use
[21:59] cases and designed with our application
[22:03] requirements in mind.
[22:07] That's all I had. Olga,
[22:09] thank you very much. there was a quite
[22:10] um rough overview and to kick off you
[22:13] know the conversation and to give them
[22:16] uh all the audience a little bit of a
[22:17] moment to uh get their thoughts together
[22:20] for questions. So you've mentioned quite
[22:22] a few applications where do you feel
[22:25] that okay in which because you do n
[22:28] orchards and turf and also construction
[22:30] where in which maybe particular
[22:32] application is now currently the most
[22:35] challenging one and where exactly the 3D
[22:38] sensors fall short there
[22:40] that's a great question I I think
[22:42] everything that I've talked about you
[22:44] know is is now right um certainly
[22:49] there's a lot more applications that are
[22:51] that are under development that I'm that
[22:52] I'm not going to speak about here in
[22:54] this forum and they all have their their
[22:57] challenges right uh even the products
[23:00] that we um that we have uh in the market
[23:04] but I would say just in general you know
[23:06] the ability to operate in all
[23:09] environmental conditions right is um I
[23:13] would say a common thread between uh all
[23:16] of the applications
[23:18] if we think you know if we meet with
[23:20] Exactly with the same meeting in three
[23:22] five years. What do you think will be
[23:24] what are your shopping list of the
[23:26] improvements E3D sensing techniques for
[23:28] these three to five years? You've
[23:30] mentioned that in the last two they
[23:31] already have done a lot and you're
[23:33] optimistic about capabilities of MC FMCW
[23:36] or longer wavelength. So for the next
[23:38] three to five years your shopping list
[23:40] of wishes where does the 3D sensing has
[23:43] to improve?
[23:44] Yeah. Yeah. Uh absolutely. Uh again, I'm
[23:47] going to reiterate when it comes to uh
[23:50] image sensors,
[23:52] HDR, global shutter has a lot of value,
[23:57] right? And um I think that's an
[24:00] overlooked uh opportunity
[24:03] uh for for image sensor manufacturers
[24:06] that has has not been adequately
[24:08] exploited when it comes to LAR. uh
[24:12] really really
[24:15] excited I I would say again about the
[24:17] FMCW LAR and the potential impact of
[24:20] integrated
[24:23] the uh ability to scale these LARs and
[24:26] uh drive the unit cost down uh to
[24:29] something that is uh more palatable for
[24:33] OEMs uh like John Deere.
[24:35] So pretty excited about that.
[24:38] Um and then you know there is this space
[24:42] sort of in between
[24:45] uh LAR and radar you know uh and I think
[24:47] maybe you alluded to that uh a little
[24:49] bit in the intro um you know interesting
[24:53] to see what happens there in terms of
[24:55] exploring that space uh in between
[25:00] um LAR and radar. Mhm. I actually one
[25:04] one last question from my side before we
[25:06] go to the audience because I'm really
[25:07] curious because you know I always when
[25:09] you see one thing when you see a car and
[25:12] you think that okay you know a little
[25:13] bit the price range of the automotive
[25:15] but when you see one of your machines
[25:18] the the uh let's say the entire
[25:19] equipment is much more expensive. Are
[25:22] you still um would are you still the
[25:25] same uh exactly sensitive to the really
[25:28] to the price point of the sensing
[25:30] technology as the automotive or do you
[25:33] allow yourself a little bit more
[25:34] freedom?
[25:35] Well, certainly. Yeah. Yeah. I mean, I I
[25:38] would say in terms of what we can
[25:40] afford, it's probably, you know,
[25:41] proportional to the cost of our uh
[25:44] equipment, but I would also say that the
[25:46] price of a lot of these components has
[25:49] still been prohibitive uh compared to
[25:52] the value uh that the customer is
[25:55] looking for, right? Uh in their
[25:57] solution, right?
[25:58] So, there's only so much that we can
[26:00] charge for these solutions, right,
[26:03] before they no longer provide value to
[26:05] the customer. So um yeah there there are
[26:09] certain price targets that we have to
[26:11] hit uh in order for these these things
[26:13] to make sense.
[26:15] Thank you for your comment. We have and
[26:16] I will call now all the questions from
[26:18] the audience. Uh from the Chilas Mitri
[26:21] you have a question about the uh for the
[26:24] spectroscopic.
[26:25] Thank you. Thanks for the very nice
[26:27] presentation. It's really impressive of
[26:29] what kind of imaging systems you have
[26:31] put on all these machines here in the
[26:35] Netherlands. They also are very playful
[26:38] with these toys. Uh so I quite often see
[26:41] them playing in the fields here in a
[26:43] tren area. Yeah, there's a lot of
[26:45] agriculture.
[26:46] Yes, the farmers there are very
[26:47] progressive.
[26:49] It's impressive. It's really nice to see
[26:51] this high-tech. Really nice. I was
[26:54] curious. I recently had
[26:56] uh with another partner an idea of for
[27:01] example more spectroscopic uh imaging
[27:04] systems that you do some some bug
[27:07] detection there are going on or some uh
[27:10] there's a lot of emission regulation
[27:12] going on there that you do do monitoring
[27:14] what what the emissions are. Uh so they
[27:16] need all special colors of lasers of
[27:19] light sources to to address some uh
[27:22] spectroscopic features in the field. I
[27:24] is there anything going on in in your
[27:27] field that you have to do some emission
[27:29] control, bug repellent or
[27:33] pesticides concentration measurements in
[27:35] the field while you're on the fields?
[27:39] Um,
[27:40] no, but but we do use um you know
[27:45] spectroscopic uh sensors uh quite a bit
[27:48] on our uh equipment. Uh people will
[27:51] probably be shocked to hear that for
[27:54] over 20 years. We've had a near infrared
[27:57] spectrometer uh offering uh that goes on
[28:01] our self-propelled forage harvesters to
[28:03] measure things like
[28:05] um uh moisture level uh protein starch
[28:09] content,
[28:11] uh fiber content, things of that nature.
[28:14] Um so yes, we're we're always very
[28:16] interested in measuring
[28:19] properties of the crop uh really
[28:21] throughout uh the growing cycle. you
[28:23] know, those are valuable uh agonomic
[28:26] insights. It does get a little bit
[28:28] challenging though when uh you're maybe
[28:31] not able to control the uh uh the
[28:35] ambient lighting, right? Um in terms of
[28:38] the utility of those uh spect
[28:42] spectroscopic systems, but in those
[28:43] scenarios where we can control the
[28:45] lighting, uh there's there's often quite
[28:47] a bit of value.
[28:48] Okay. Oh, good to know. Thanks.
[28:52] Thank you, Mitri. you will be soon on on
[28:54] the presentation floor. But but the at
[28:56] the moment we have another question from
[28:58] the audience. Agusto Himenez from
[29:00] Cognissi. Austo you have a question and
[29:03] also a comment. So maybe please unmute
[29:05] yourself, introduce what Cogni actually
[29:07] does and ask your question
[29:12] Austo.
[29:15] I'm unable to speak. Sorry. Okay then I
[29:19] will read it out. So DT DOF um is more
[29:23] robust to skater in media that f than
[29:25] FMCW. FMCW loses coherence very quickly.
[29:30] What kind of resolution and range would
[29:32] be required for John Deere.
[29:34] Let me just say that
[29:37] what he said about FMCW and DE versus
[29:40] DTO does not match my experience. Um the
[29:45] second thing I'll say is that uh you
[29:48] know we have a lots of different
[29:49] applications with a broad spectrum of
[29:53] requirements right uh in terms of uh
[29:57] accuracy and resolution.
[30:00] Um I would say for the vast majority of
[30:03] applications the more resolution uh the
[30:05] the marrier right um but for especially
[30:10] like in autonomous type um applications
[30:14] uh we do have other you know automation
[30:16] applications though where uh you know
[30:19] we're looking for extreme accuracy uh I
[30:22] would say you know in the
[30:26] you know centimeter range right in terms
[30:28] of of absolute accuracy for for range.
[30:32] Um yeah, I would say that's probably our
[30:34] most stringent requirement.
[30:39] Um August, you were you happy with an
[30:42] answer?
[30:43] But Zach, you just said that that DTO is
[30:47] not more robust to scattering media than
[30:50] MCW
[30:53] direct time of flight.
[30:54] I would say just in terms of the the
[30:56] ability to to penetrate uh
[31:00] the dust that we observe in our
[31:02] environment.
[31:04] Are are you using
[31:05] that's probably now now how much of that
[31:06] is a function of
[31:09] you know um DTO versus FMCW
[31:14] I don't know it's probably more of a
[31:16] function of the wavelength
[31:19] but uh yeah
[31:22] I think we are comparing of course
[31:24] apples and oranges but I think the the
[31:26] killer the killer question here is you
[31:28] said very clear that lighter is just not
[31:30] good enough when you have dust um and I
[31:33] like that very much. The statement is
[31:35] very much clear to me. But what what is
[31:39] the solution and most important what do
[31:42] these people in the room need to do to
[31:44] satisfy a solution here?
[31:47] Well, I would say you know um every bit
[31:51] of improvement in dust penetration adds
[31:54] value,
[31:57] right?
[31:58] So um
[32:01] you know certainly as the wavelengths
[32:03] get longer right we would expect to see
[32:05] better and better and better uh dust
[32:08] penetration.
[32:09] Mhm.
[32:10] Uh the other thing I would I would say
[32:13] you know is uh there is a a large
[32:17] spectrum
[32:18] right uh in between LAR and radar like I
[32:22] said that is is
[32:24] largely unexplored shall we say. Um I I
[32:27] think we're starting to see more
[32:29] exploration of that space. Uh and that
[32:32] might get us to
[32:35] um a place where you know we have at the
[32:39] same time you know uh the resolution
[32:41] that we need uh as well as the dust
[32:44] penetration.
[32:45] Henrik Matson is the CEO of SPIO and has
[32:47] exactly the same question I had in mind
[32:49] but I think the question will sound much
[32:51] better if it comes from him.
[32:53] Yeah Henry. Um hello uh Zachary um uh
[32:58] yeah you you're teasing for longer
[32:59] wavelength. Can you reveal some some
[33:01] range of the wavelength for that?
[33:06] Um wavelength for what exactly?
[33:08] Uh the the measurement the the distance
[33:11] measurement. Yeah.
[33:13] Um
[33:15] I can also rephrase the question in a
[33:17] different way. what kind of particles
[33:18] size of the particles you're dealing
[33:20] with because that's going to define more
[33:22] or less what wavelength it's going to go
[33:24] through.
[33:24] Yep. Yep. Uh absolutely absolutely. So
[33:29] um
[33:31] yeah, I don't have my my numbers with me
[33:35] right now, but I um my recollection in
[33:38] terms of you know kind of coarse dust
[33:41] particles, you know, tend to be maybe in
[33:44] the 2 mm range.
[33:48] Right. Uh that would be kind of
[33:50] extremely coarse uh dust particles.
[33:59] Yeah. So, so that's that's really what
[34:00] we're yeah probably the worst case
[34:02] scenario that we're that we're trying to
[34:04] penetrate.
[34:05] Okay. I also would think because um I
[34:09] mean I will throw this word into the auh
[34:11] into the discussion because it's thrown
[34:13] into any discussion nowadays because a
[34:15] lot of the sensing now is also starts to
[34:17] improve the sensitivity or the quality
[34:19] of the data not because of the sensor
[34:22] but because of the data processing and
[34:24] how AI basically or machine learning
[34:26] let's generally say is able now to
[34:29] interpret the data. Are you also looking
[34:31] into this direction? Maybe let's stop to
[34:33] improve the sensors and let's start
[34:36] think more carefully what we actually do
[34:38] with the data that we harvest.
[34:40] That that is really an excellent
[34:42] question um Olga and certainly we've
[34:46] been I would say very much on the
[34:48] cutting edge of deploying machine
[34:50] learning uh on our equipment. Okay. um
[34:55] we any we're very very expert uh at
[34:59] that. Um but most of that
[35:03] um knowhow and experience is really uh
[35:07] related to to image sensing. Um but I'm
[35:11] also very excited about the potential of
[35:14] machine learning to be applied to uh
[35:17] radar signals and potentially LAR
[35:19] signals uh as well especially the more
[35:23] like the raw waveforms um essentially
[35:26] right I think the raer the data the
[35:29] better uh the AI is going to be able to
[35:33] uh coax out uh
[35:36] insights right that we can't get from
[35:39] conventional signal processing.
[35:42] So that is also a direction, you know,
[35:44] I'd like to see some of these sensor
[35:46] technologies go. Uh so make make a
[35:49] really good uh quality sensor and then
[35:53] uh find a way to expose those raw
[35:56] sampled waveforms uh through your
[35:58] interface um you know over over high
[36:01] bandwidth interfaces.
[36:04] Okay. I think there was uh then this was
[36:06] for me also the um an interesting
[36:09] transition to introduce the next speaker
[36:11] from
[36:12] point Aust
[36:17] thank you so much and sorry for not
[36:19] being able to speak earlier but my
[36:21] comment about the FMCW versus the direct
[36:24] time flight is uh I mean with respect to
[36:27] the um to the wavelengths itself they
[36:29] are very close I mean FMCW sometimes
[36:32] people are operating at 1550 50 and
[36:35] light that operates at that wavelength
[36:36] as well. But at that kind of between
[36:39] either near infrared to shortwave
[36:42] infrared there's very little difference
[36:43] in the in the wavelength. So from the
[36:45] particle perspective and the signal
[36:48] penetration perspective I think they are
[36:50] similar. Uh my comment about the being
[36:53] more robust is mostly about coherence.
[36:55] FMCW relies on coherence. it's very
[36:58] difficult to get maintain the coherence
[37:01] with targets already in a clear day uh
[37:04] and then with dust and all the um you
[37:07] know scattering from the media that's
[37:09] that's what my comment was about and it
[37:11] is just rely on the signal penetration
[37:13] itself um but I think the comment is
[37:16] mostly also looking into uh the way to
[37:20] expose the scene typically lighters
[37:23] especially automotive lighters the
[37:25] spinning lighters u you have um very few
[37:29] laser pulses per position. So the
[37:30] spinning is very quick. So your chances
[37:33] of collision those the light collision
[37:35] and losing that reflection is very high.
[37:37] Um and I was my comment was mostly about
[37:41] different operations maybe global
[37:42] shutter flash mode operation in dto
[37:45] longer exposure many more uh laser
[37:47] pulses you actually be able to um uh to
[37:51] cope with that scattering media. And I
[37:54] I'm I'm wondering um technologies like
[37:56] that is something that um Cogniz is
[37:58] working on. It's something that doesn't
[38:00] exist today. You know, non scanning in a
[38:03] high resolution uh and I'm wondering you
[38:06] know about the resolution and range that
[38:09] uh uh John Deere is looking for and I I
[38:12] understand your your your answer.
[38:14] Yeah. Yeah. Um range is probably easier
[38:17] to answer, right? We don't need
[38:18] extremely long range. uh most of our
[38:20] applications,
[38:22] you know, 50 meters or less is is really
[38:26] what we're looking for. Nothing compared
[38:27] to, you know, what would be required for
[38:30] for automotive, I would say. Um again,
[38:34] resolution would be highly variable.
[38:36] Now, I mean, one thing that we could do,
[38:38] right, is we've uh we've actually
[38:41] constructed a dust chamber
[38:43] at our facility in Fargo. And um if you
[38:48] have a
[38:50] uh theory uh or uh a system uh that you
[38:55] want to test, let's uh let's talk about
[38:57] it and uh you know do a study in our
[39:01] dust chamber.
[39:02] That sounds great. Yeah,
[39:03] Zah, I remember that was the same
[39:05] invitation to uh Luminar was it two
[39:07] years ago. We have a dust chamber. Come
[39:10] and try.
[39:12] Yeah, absolutely.
[39:13] Austo, what what does Cogncy do? A
[39:15] couple of words about yourself.
[39:17] Yeah, so Cognc is developing a in pixel
[39:21] his uh in pixel signal processing. So
[39:23] histogramming on the pixel. So basically
[39:26] a miniaturization of a lighter
[39:28] technology in a solid state in within
[39:30] each of those pixels. So pixel can u
[39:33] range from uh it can cover from 10 to 50
[39:37] meters as um as the requirements the
[39:39] typical requirements for mid-range. And
[39:41] what we usually do is that this pixel is
[39:44] very small as in 10 micrometer pitch uh
[39:47] and it can produce the full resolution
[39:49] or full uh histogram within the pixel
[39:52] itself. So we can create arrays that are
[39:54] small for you know very very small
[39:55] applications or very small resolutions
[39:57] or even going to the multi- megapixel um
[40:00] resolutions all depending on the power
[40:02] consumption and so on. So really
[40:04] bringing the um direct time of flight
[40:07] into solid state very compact in flash
[40:10] modes is what we are uh targeting.
[40:13] Okay, thank you so much.
[40:16] We have a question in YouTube. So allow
[40:18] me to a question from Sakonas. This
[40:20] meeting is live streaming YouTube. So
[40:22] those YouTubers in the world please
[40:23] write your questions in the chat. Zack
[40:25] the question is from Simon and he's
[40:28] interested in thermal cameras. They say
[40:30] that they have seen huge interest in
[40:32] automotive for new applications of
[40:35] thermal cameras lately and they're
[40:36] wondering if you have seen the same and
[40:39] there are certain business cases for new
[40:42] applications in thermal cameras.
[40:45] Yeah. Yeah. Certainly um you know
[40:48] thermal is is something that uh we're
[40:52] we're interested in uh as well. I think
[40:55] the key for thermal is going to be that
[40:59] uh sort of like the temperature
[41:00] resolution right of the thermal cameras
[41:03] in our application space uh that's going
[41:06] to be the most important quality of of
[41:10] the thermal sensor
[41:14] and with the so much basically data
[41:17] being produced uh and depending
[41:20] independent of the sensing technique do
[41:21] you do all the uh manipulation and data
[41:24] processing on board of the of the
[41:26] equipment post-processing or how is it a
[41:30] life measurement or you do do you do
[41:32] process the data later?
[41:34] Oh yeah, a very good question. Um most
[41:38] of the systems that I talked about here
[41:40] today really almost all of our
[41:42] perception sensing systems are are part
[41:45] of some sort of control system or uh you
[41:48] know real time feedback loop. So the
[41:51] processing needs to be done uh on the
[41:53] machine almost always.
[41:56] Thank you. And I would like with this uh
[41:58] little bridge I would like to introduce
[42:00] the next speaker which who is the Ramos
[42:02] Nicolesco from point cloud who is able
[42:05] potentially to help you to extract much
[42:07] more useful information from all the
[42:09] data that you are collecting. Ramos you
[42:12] are on stage.
[42:14] Thank you. So I think I'm just going to
[42:17] go quickly through sort of the
[42:20] introduction of the company. So we're uh
[42:24] we're a company we've actually started
[42:26] we've we've been around for a very long
[42:28] time. So we've started about 10 years
[42:29] ago. Uh we pioneered a new technology
[42:32] which is uh these days is I think most
[42:35] people there are a lot of names people
[42:37] refer to in the literature but we
[42:39] usually use coherent focal plane array
[42:41] technology. Some people use switched
[42:43] arrays or all sorts of other things. But
[42:47] um at the end of it, it's we we
[42:50] manufacture these chips which are the
[42:53] easiest way to think about it. It's like
[42:54] a it looks and feels like an image
[42:56] sensor. It's a large array
[42:57] two-dimensional array of of uh pixels
[43:01] and you can think of it as each pixel as
[43:03] being a coherent transceiver. Um
[43:07] we uh essentially by doing this we we
[43:10] combine two kind of old concepts. One is
[43:12] coherent ranging that comes from radar
[43:15] been around for a long time with focal
[43:17] plane array architecture. So what you
[43:19] get is the best of both worlds. You get
[43:22] all the advantages of coherent ranging
[43:24] um resolution range accuracy velocity
[43:27] measurement everything that Zach
[43:29] mentioned earlier that they would like
[43:30] to have. Um and it is a focal plane
[43:35] range sensor. So it's just a chip in the
[43:37] focal plane of a lens. So it's very
[43:39] similar to a camera as far as
[43:40] architecture. Um it's very simple from a
[43:44] from a system integrator perspective.
[43:46] It's a one chip, one lens. There are
[43:48] there is no complex alignment. It uh it
[43:51] it um
[43:53] relies on reciprocity of light
[43:55] transmission. Light comes out through
[43:56] the pixel and comes back to the pixel uh
[43:59] naturally.
[44:01] Um we use silicon camos manufacturing
[44:04] the standard process from global
[44:06] foundaries. Everything is solid state.
[44:08] There are no moving parts. um sensor is
[44:11] just as robust as a regular camera. Um I
[44:16] will go to the picture to the right
[44:17] which is uh showing this is a real point
[44:21] cloud. It's not a simulation or anything
[44:23] like this. It's sort of a partial team
[44:26] picture the guys in the in the lounge
[44:29] area of our office in in Zurich.
[44:32] It's a it's taken with a
[44:35] a sensor. It's a near QBGA. It's a 352 x
[44:39] 176 pixel sensor. It's on a gen one EVK.
[44:42] Um, what you see there, it's about it's
[44:45] a field of view of 33 degrees by 20°
[44:48] with an angular resolution 08. Uh, this
[44:51] is the the field of view and angular
[44:54] resolution is really dependent just like
[44:55] in a camera on the lens you choose. We
[44:59] um if you go on our website, we actually
[45:02] recently published this sensor in a in a
[45:05] nature paper uh two two months ago. You
[45:08] have all the details of uh of um
[45:12] the sensor as well as uh as
[45:14] characterization with different lenses.
[45:16] You can just stick different lenses on
[45:18] it and it'll just work like a camera
[45:20] typically. So you get wider field of
[45:22] view with less angular resolution or
[45:25] narrow field of view with more
[45:26] resolution. So in the middle it's um
[45:31] sort of a list of capabilities. So while
[45:34] this sensor was designed for mid-range,
[45:35] this picture is taken at about 11 12 m
[45:38] which we would consider very short
[45:39] range. Uh the sensor was divi designed
[45:41] for 65 to 70 m. It actually performs
[45:46] reasonably well up to about 4050. We
[45:49] have taken point clouds with it all the
[45:51] way to 125 m. Uh having said that we
[45:55] don't recommend it past 50 meters just
[45:58] because the probability of detection
[46:00] degrades a little bit and u also false
[46:02] positives are getting a little bit
[46:03] higher than what is reasonable for a
[46:06] commercial application. So um also what
[46:09] I would mention this is a first
[46:12] iteration at scale. So this is not
[46:13] actually a sensor. We we sample some
[46:15] sensors. If anyone is interested to buy
[46:17] an evaluation kit uh we do have them
[46:19] available for sale. Uh this sensor is
[46:22] not going to go into volume production.
[46:24] It's u it's only a prototype an
[46:27] engineering sample and uh currently uh
[46:31] the next generation is going to
[46:32] manufacturing which is a standard QVGA.
[46:35] It's a 320x 240.
[46:37] Um it will have much longer range. It
[46:41] will be configurable to go anywhere from
[46:44] 60 to 70 mters which we would consider
[46:46] short range to uh close to 200 m which
[46:49] would be the long range. Um as far as I
[46:54] mentioned there are capabilities. So in
[46:56] general up to now we've been working on
[46:58] an ASIC model in the sense that this
[47:00] particular chip that you see the in the
[47:03] pictures there the the the bigger one
[47:06] was designed at the on the specs of a
[47:09] customer. So it's a automotive customer
[47:12] that actually paid for the for the
[47:14] development.
[47:16] Um we in general the capabilities we
[47:20] have right now we we based on existing
[47:23] silicon performing it in different
[47:25] configurations. We we can do anything
[47:27] for from short range to possibly as high
[47:30] as 400 meters uh resolution anywhere up
[47:34] to about VGA with a current pixel size.
[47:37] I would say that not all of these are
[47:39] simultaneous. So please keep that in
[47:41] mind. Um, we're not going to tell you
[47:43] that you can get all of those five
[47:45] things in the same time. Uh, we do can
[47:48] get we can get you high depth precision
[47:51] essentially depth noise at as low as if
[47:54] needed even below a millimeter. We
[47:56] generally never had anyone asking for
[47:58] that, but if you need it, it can be
[48:00] done. Uh, radial velocity as low as a
[48:03] millimeter per second. Um,
[48:06] and power. Again, as I mentioned, this
[48:08] is not for a VGA type sensor. That would
[48:12] be more like in the 10 watts plus range.
[48:14] But if somebody wants a small sensor for
[48:17] mobile, for consumer applications, we do
[48:19] have configurations on which we can go
[48:21] below half a watt.
[48:25] Um,
[48:28] this is a page. This is actually at the
[48:30] bottom. It's like a snapshot from sort
[48:33] of the basics how this works. I was
[48:34] mentioning maybe I'll I'll go through
[48:36] the pixel. So at the left is the pixel.
[48:39] Um for people that have a little bit of
[48:41] an optics background everything is done
[48:43] in silicon photonics. So you see uh in
[48:46] green are our silicon waveguides and we
[48:49] split the light from a bus and we send
[48:51] it towards two grading couplers towards
[48:53] the outside world. It couples back to
[48:56] the same after it bounces off the target
[48:57] it couples back to the same gradings. It
[49:00] goes in the reverse direction and goes
[49:02] into the second there are two detectors
[49:05] there two German germanmanium or silicon
[49:06] detectors where it gets mixed with local
[49:08] oscillator and it's a it's a coherent
[49:11] receiver and then there is a transistor
[49:13] amplifier on each pixel and now once you
[49:16] make a pixel you can see here a block uh
[49:19] that
[49:21] is is doing the imaging we generally
[49:24] illuminate a row or or a column
[49:28] depending of how the sensor is
[49:30] configured uh at the same time. So it's
[49:32] a little bit like rolling shutter here.
[49:35] Uh we illuminate a row and then we read
[49:37] it. Then we move to the next row, we
[49:39] read it and so on until we finish the
[49:41] frame. Now sort of coming back to the
[49:44] high level
[49:46] um you can think of this as we
[49:48] essentially enable a camera like system
[49:50] architecture. So it's a kind of like the
[49:52] equivalent of a SIMOS image sensor for
[49:54] the fourdimensional world. It is
[49:56] coherent. So that's the fourth
[49:58] dimension. So we can extract radial
[49:59] velocity as well. Um it does have all
[50:03] the advantages of coherent technology
[50:04] performance and it has the the
[50:06] scalability of a focal plane array
[50:08] architecture.
[50:11] Um everything is done in large scale
[50:15] production quality um processes. So we
[50:18] work with global or front- end
[50:20] manufacturing. We work with packagers
[50:23] both small and large. But we do have a
[50:25] also already working with a volume
[50:28] packager. Um and um yeah in general it's
[50:33] it's everything is done so it can be uh
[50:37] going into production and also have a
[50:40] compelling cost or price point uh high
[50:44] performance and lower cost and uh at
[50:46] this point we actually think that we do
[50:48] have we're generally benchmarking
[50:49] against Sony's the IMX 479 and so um we
[50:54] do think that at this point we have a
[50:56] lower cost than competing DTO sensors
[50:58] based on just area of silicon and then
[51:00] complexity of the process and in some
[51:03] ways we're only getting started so we
[51:04] think there is a lot of room to provide
[51:07] this level of performance at a very very
[51:09] compelling cost u and also I would point
[51:12] out here because of the monostatic
[51:13] architecture we save the system
[51:16] integrators a lot of a lot of u of um
[51:20] parts that otherwise and alignment steps
[51:23] and all sorts of other things which are
[51:25] a little bit painful in manufacturing
[51:27] Um as far as I already mentioned that as
[51:30] far as uh we can do short range to over
[51:33] 200 mters. Um
[51:35] we can do large sensors if you really
[51:37] are interested and also you have to pay
[51:40] for it. It's generally large sensors
[51:43] means also a larger silicon just like in
[51:45] a camera sensor larger amount of
[51:47] silicon. Um, of course, we have all the
[51:50] advances of coherent immunity, ambient
[51:52] light, optical noise, and really the way
[51:55] we uh and we're actually I mean I this
[51:58] is a very very short two-page
[52:00] presentation, but we're not only looking
[52:02] at the uh even our chip. So, if you look
[52:06] at the paper, it has a lot of
[52:10] electronics integrated co-integrated
[52:11] with the optics. So, we're as you
[52:13] probably see here, there is a transpire
[52:15] on the pixel, but that's not that's only
[52:17] the beginning. There are amplifiers
[52:19] everywhere on the readout. Everything is
[52:20] fully integrated. Uh the switches which
[52:23] are being used to direct light to
[52:24] different parts of the chip. They have
[52:25] all the drivers. Everything all the
[52:27] electronics is on chip. The chip
[52:29] communicates with the digital digital uh
[52:32] uh uh communication interface a digital
[52:35] an SPI interface and uh we're kind of
[52:39] again like we're only we're we're
[52:40] integrating more and more parts into the
[52:43] chip. So we're trying to the way we look
[52:46] at it from a as far as cost reduction we
[52:50] are doing it the sort of the electronics
[52:54] the the right way we're doing it the
[52:56] right way the right way to make sure
[52:58] that this technology has a higher
[53:00] performance and lower cost compared to
[53:01] direct time of flight which is what we
[53:04] all want to do before I want before I
[53:06] give you the floor to sack so he can
[53:07] give us his his opinion and his room for
[53:11] collaboration with your remos there's a
[53:13] question in the room coming from a
[53:14] company that is very dear for me.
[53:16] Lumotive is represented today in the
[53:18] room by Apurva Jane. Apurva, what's on
[53:20] your mind?
[53:23] All right. Hi Re is a very very good
[53:25] presentation. I think it's a very
[53:26] exciting technology and I asked my
[53:28] question before I saw the second slide.
[53:30] So let me just maybe rephrase my
[53:32] question a bit. uh one I wanted to
[53:34] understand what type of transmitters uh
[53:36] is your technology or or this focal
[53:40] plane array compatible with and when you
[53:42] compare the cost or performance are you
[53:45] comparing it for the full system or just
[53:47] the chip to chip right so to build a
[53:49] full LAR system you need to have the
[53:50] transmitters and things I have a second
[53:52] question as well rea sorry I'll just ask
[53:54] it together so you can answer it and
[53:56] then where do you see the scalability of
[53:58] the technology because it's very
[53:59] exciting where is the technology
[54:00] scalability for field of too, right? And
[54:04] frame rates like what are your uh you
[54:06] know where do you see this headed with
[54:08] with with this type of technology?
[54:10] That's a lot of questions. So um we'll
[54:12] we'll try to address one by one and if I
[54:14] forget you can kind of get me back on
[54:16] track.
[54:17] Uh so first the transmitter. So we
[54:19] actually look at it. So first of all we
[54:21] are a tier one tier two supplier. So
[54:23] we're not building systems. We are
[54:24] building an EVK. So it's it's a
[54:26] reference design like any other
[54:27] semiconductor company. But we are a
[54:29] semiconductor company. we have no
[54:31] intention to uh compete with like I
[54:33] don't know iva or any of the tier ones
[54:36] or any of that. So for us those are
[54:39] customers. So um that's one. Second you
[54:42] asked about transmitter. So we're
[54:44] currently using an external modulation
[54:46] scheme. Uh we are not married to a
[54:50] particular transmitter. So there are a
[54:53] lot of people in the industry do direct
[54:54] modulation of the laser. The main reason
[54:56] why we haven't done that and we continue
[54:58] for now to do external modulation. So we
[55:01] use an inface quad modulator which is
[55:04] built in also in sleeponics is the same
[55:05] process and it's actually going to be
[55:08] integrated with the array. Uh we're also
[55:11] going to integrate the driver. So as I
[55:14] said we're trying to bring more and more
[55:16] all the chips that the customer or the
[55:19] system integrator we're selling to is
[55:21] normally has to buy from someone else to
[55:22] put them on the board. we try to already
[55:25] kind of integrate as much as possible so
[55:26] that way we reduce the overall bomb for
[55:29] the system integrator. That's kind of
[55:30] how we think about it. So anyway, in
[55:33] general, anything that we feel like
[55:36] needs to be tailored or it would save
[55:39] cost if we bring it on chip, we're going
[55:41] to bring it on chip from the system
[55:43] level.
[55:44] Yeah, we have very similar approach as
[55:45] well like because we're coming from the
[55:47] uh you know beam sharing side. So in
[55:48] this case when you're comparing kind of
[55:50] the DTO versus this approach you're
[55:52] looking at the system level not just the
[55:53] chip level. Is that right?
[55:55] Well we're looking at the system level
[55:57] only from the perspective of if a part
[56:00] that is on the board we can bring it in.
[56:05] It's relatively simple and it will save
[56:08] the customer
[56:10] a dollar or two and we'll bring it in.
[56:14] If it's something is just commodity,
[56:15] won't save any money, we let the
[56:17] customer just buy it from someone else.
[56:19] But we look at it from that perspective.
[56:20] If it makes sense, we'll bring it in and
[56:23] it saves money. Uh if not, we leave it
[56:26] alone. Now, coming back to your
[56:27] question, we're not married to external
[56:29] modulation. We've been using external
[56:31] modulation because it actually works.
[56:33] It's extremely robust. And what people
[56:36] quite often don't understand is
[56:39] there are a lot of direct modulation
[56:42] systems out there. Um if your direct
[56:46] modulation system is not super linear,
[56:48] you're going to take a very serious hit
[56:51] in signal to noise ratio. Uh it is very
[56:54] important for your transmitter to be
[56:56] very very linear
[56:58] and you can offset that by just throwing
[57:01] more power at the problem. uh what we're
[57:03] seeing now while while that was okay at
[57:06] the beginning of LAR today is not okay.
[57:09] Our customers are asking us how is your
[57:11] power consumption versus what Hesypus is
[57:13] producing. So if I throw an extra five
[57:16] watts that's not okay. Um so I would say
[57:20] that we're sticking right now with
[57:22] direct with with external modulation
[57:25] mostly because it works. It's extremely
[57:28] efficient. It's extremely clean. It's a
[57:30] more complex than direct. We haven't
[57:33] seen yet a direct transmitter that we're
[57:35] convinced it will work without
[57:37] increasing the power budget.
[57:40] You're doing a fantastic job and you're
[57:41] going to hear from Apurva in a second,
[57:42] but let me give the floor to Henrik
[57:44] Matson from SPIO who has a very good
[57:46] question. Henrik.
[57:48] Yeah.
[57:49] Uh hello. I I saw your schematic setup.
[57:52] Uh would it be beneficial to have some
[57:54] kind of a lens system that can that can
[57:58] concentrate the the the signal to to to
[58:00] to the graings?
[58:02] Sorry, I didn't catch that.
[58:04] Um yeah, see you have a graating
[58:06] couplers to to waveguards. Those grating
[58:08] couplers are not the most efficient in
[58:13] the world. Would it be beneficial to
[58:15] have some lenses that can deliver the
[58:17] beam in concentrated form to to the
[58:19] braing so you can have more signal? Um
[58:21] okay so it's actually they're more
[58:24] efficient than you think. So let me just
[58:26] first answer the uh so also if you look
[58:29] at the paper we do have a a lens there
[58:31] but for complete different reason. So uh
[58:34] the grading is actually light is
[58:36] transmitted through the grading. So it's
[58:38] coming out of the grading is coming back
[58:40] into the grading. Uh grading uh loss is
[58:44] um between one and a half and 2 and a
[58:47] half dB roughly. uh and this is free
[58:51] space. So you have to have in mind that
[58:53] this is not the same as coupling from a
[58:56] gradient to a fiber because we don't
[58:57] need to modatch. We just need to keep it
[58:59] clean. So that's number one. Uh second,
[59:03] we are using a microl lens. We are using
[59:05] a micro lens for a very different okay
[59:06] on the way back. Maybe I should cover
[59:08] that first. Um in coherent you're only
[59:12] coupling a single transverse mode. So
[59:14] you can think of it as a laser beam. So
[59:18] it doesn't you don't you're not actually
[59:20] by collecting more light you're actually
[59:22] not doing anything you're not improving
[59:25] your SNR because it's incoherent. So and
[59:29] that scheme works in a biatic because
[59:31] there you have separate lenses. So you
[59:32] technically could collect more light but
[59:34] it won't couple into the grading because
[59:38] it the only thing that matters is what's
[59:40] in the trans in the fundamental mode. I
[59:43] don't know. This is a it's a bigger
[59:44] question and it's a it's a bigger
[59:46] discussion.
[59:46] It's a very good question. But I can
[59:47] tell you one thing from the bottom of my
[59:49] heart. Evaluate the the optics layers of
[59:52] SPIO for encoupling of the light. You'll
[59:54] be surprised because the great that you
[59:57] have a very narrow
[59:59] language selective and these ones have
[01:00:02] potential to to do some things as well.
[01:00:04] But I would like to leave it there
[01:00:05] because now what I want to do is to go
[01:00:07] to Zach Bonafas, show him this slide and
[01:00:10] ask him is this useful to you in any way
[01:00:14] and what kind of feedback would you give
[01:00:15] to Remos to point cloud sack bonafas?
[01:00:18] Yeah. Uh Remis uh thanks for for the
[01:00:21] presentation. Uh very very interesting
[01:00:23] technology. Um certainly would would
[01:00:26] love to follow up um after this meeting.
[01:00:30] Uh maybe a couple questions for you. So
[01:00:33] um you know is there a particular
[01:00:36] wavelength that this uh technology
[01:00:38] operates at or is that variable?
[01:00:41] So we're currently working at 1 point
[01:00:43] this this sensor is working at 1.3
[01:00:45] microns. So it's in the O band. We have
[01:00:47] done in the past sensors at 1.5. We kind
[01:00:51] of switched away from it because we feel
[01:00:52] like 1.3 is a better compromise from a
[01:00:55] number of reasons. I won't go into it
[01:00:57] right now because there are about 10
[01:00:58] different reasons why
[01:01:00] pro pros and cons of the digital
[01:01:02] wavelengths and we could go into it. By
[01:01:04] the way, we did I I had a question a
[01:01:07] little bit for you because I was like we
[01:01:09] had conversations before with some of
[01:01:10] your customers
[01:01:12] and they mentioned dust and we haven't
[01:01:16] had the chance. We tested it in fog. It
[01:01:19] penetrates as long as we have enough
[01:01:22] power to penetrate through the fog. I'm
[01:01:23] assuming that's pretty much the case in
[01:01:25] dust. I would also assume that we're
[01:01:27] doing okay because of uh there is no
[01:01:29] blinding effect here. So in dto I think
[01:01:31] on dust you're probably going to blind
[01:01:32] the sensor but you know better.
[01:01:35] One way to find out right uh
[01:01:37] yeah we haven't tested it off to see how
[01:01:40] but we would expect that it's going to
[01:01:42] blind the sensor because you have some
[01:01:44] text scattering that is going to start
[01:01:47] getting
[01:01:48] it has said that he's going to follow up
[01:01:50] with you after the meeting. That is why
[01:01:52] we are here. That is why we're here. I'm
[01:01:54] very happy and I'm also very happy
[01:01:57] company that I'm going to introduce now.
[01:01:59] It's a company that means a lot to me as
[01:02:01] you know because you see my linking I go
[01:02:03] to CES every year and I always try to
[01:02:06] find out the innovation the biggest
[01:02:08] innovations being there and the LCM
[01:02:10] component of Lumotive always calls for
[01:02:12] attention there are year after year
[01:02:14] always nominations of innovation awards.
[01:02:16] Lumotive is a great company and Apurva
[01:02:19] is here. Aba Yen is here to tell us also
[01:02:21] not that they are great company but also
[01:02:22] how can they address some of what they
[01:02:24] are many needs presented by Y in the
[01:02:26] beginning. Abula thank you very much
[01:02:28] join the meeting today. The floor and
[01:02:29] the attention of all my friends is
[01:02:31] yours.
[01:02:32] Always a pleasure always. Thank you Jose
[01:02:34] and thank you Olga for having me here uh
[01:02:36] again and I'm uh very excited what I'm
[01:02:39] going to do if you guys can see the
[01:02:41] presentation. I was going to present a
[01:02:42] video by the way so it would be in one
[01:02:44] slide but I'll still try to keep it
[01:02:45] within uh 5 minutes. Okay. Um I'm going
[01:02:48] to present a little bit of a systems
[01:02:50] level view, end use view because at
[01:02:52] Lumotive over the last couple years what
[01:02:54] has really changed is we've really
[01:02:56] started looking at what the end
[01:02:58] customers need you know and we're trying
[01:03:00] to separate like you know what the end
[01:03:01] customers need and we want to address
[01:03:02] that total cost of ownership and solving
[01:03:05] their problems that are actually
[01:03:07] happening at the ground at their
[01:03:09] operation level right and then we bring
[01:03:12] it back to our chip and and and and uh
[01:03:14] deliver the sensors. So I'll talk a
[01:03:16] little bit about that and I'll use a
[01:03:17] particular example and this example is
[01:03:20] related to our success in the uh
[01:03:22] robotics that's happening in industrial
[01:03:24] settings in warehouses and last mile
[01:03:26] delivery and and uh kind of the
[01:03:28] industrial process automation but the
[01:03:31] same concept extends to uh we working
[01:03:33] with you know an end user for
[01:03:35] construction equipment and they're
[01:03:37] putting the same type of sensors that
[01:03:39] that is enabled by our technology for
[01:03:41] safety sensing for 360°ree safety
[01:03:43] sensing in construction vehicles. in in
[01:03:46] Saudi Arabia desert environments, right?
[01:03:48] Um so just taking like this particular
[01:03:50] example about robots in industry and
[01:03:52] we're hearing a lot about this, right?
[01:03:53] Physic uh the physical AI layer,
[01:03:56] robotics really coming into play at
[01:03:58] warehouses and all these processes.
[01:04:00] They're becoming a common feature and
[01:04:02] the robots are going from doing single
[01:04:04] task pre-programmed to being dynamic and
[01:04:07] being in environments that are you know
[01:04:11] uh that are requiring the robots to do
[01:04:13] multiple tasks and closer to objects and
[01:04:16] closer to humans. Now this is where the
[01:04:20] previous generations of robots are not
[01:04:22] very well suitable because the robots
[01:04:25] are using multiple sensors as Zach you
[01:04:27] were also presenting like your current
[01:04:28] equipment in every direction sometimes
[01:04:31] four to six sensors to do different
[01:04:33] things. Same thing happens in the
[01:04:35] warehouse where they actually had uh two
[01:04:38] stereo cameras and two line scanners. So
[01:04:41] one line scanner looking at just the
[01:04:42] floor to know if there's a cliff or a
[01:04:44] fall or a step. They had two serial
[01:04:46] cameras for object identification,
[01:04:47] classification or collision avoidance
[01:04:49] and they had a long range uh sensor like
[01:04:52] 30 to 50 meters to look at the
[01:04:54] localization so the robot can move and
[01:04:56] imagine this in all four directions
[01:04:58] right so this is the number of sensors
[01:05:00] just 3D sensors that we're using in
[01:05:02] addition to RGB and we call this like a
[01:05:06] complex way of doing things very high
[01:05:08] cost poor scalability each sensor is
[01:05:11] purpose-built for just one type of
[01:05:12] sensing and in spite of this actually
[01:05:15] with one of the biggest uh west coast uh
[01:05:18] you know um warehouse companies you can
[01:05:20] say or e-commerce companies who is a
[01:05:22] strategic investor in Lumotive I cannot
[01:05:25] name them particularly in this case they
[01:05:28] uh had a huge operational efficiency
[01:05:31] because uh even in spite of all these
[01:05:33] sensors there was a lot of gridlocks
[01:05:34] between the robots the robots will go
[01:05:36] and catch each other and when you look
[01:05:37] at the full animation we have kind of
[01:05:39] mimicked it and they would get into
[01:05:41] gridlocks because of sensor to sensor
[01:05:42] interference they'll see too much
[01:05:44] saturation from the ground sometimes and
[01:05:46] they'll think that there's a hole in the
[01:05:47] ground and they'll stop there. Okay? Or
[01:05:49] they'll get too close to some metal
[01:05:51] structures and there's multipath uh
[01:05:53] artifacts and they wouldn't know how to
[01:05:55] get closer to those objects. So humans
[01:05:57] have to interfere uh and and unlock
[01:06:00] these uh these uh robots. There was 40%
[01:06:03] of inefficiency in these operations. So
[01:06:06] this is where Lumotive comes in with our
[01:06:08] uh chip based technology LCM beamstring
[01:06:10] technology that allows programmable
[01:06:13] sensing. With one sensor we can enable
[01:06:17] looking at the ground with the uh you
[01:06:19] know with the kind of a performance
[01:06:21] specs that you need for the ground. We
[01:06:22] are able to create a slice that is long
[01:06:24] range to look at your localization and
[01:06:26] we're able to create your region for
[01:06:29] collision avoidance all in one sensor.
[01:06:31] And this is while improving the point
[01:06:34] cloud quality, reducing the uh
[01:06:36] saturation, enabling a very high dynamic
[01:06:39] range sensing. By very high, what I mean
[01:06:41] is we do four to 5x of normal kind of
[01:06:45] laser dynamic range using our LCM
[01:06:47] technology and enable high frame rates
[01:06:50] where you need it. So with single
[01:06:52] sensor, we replaced actually four
[01:06:54] sensors. So four went to one in this
[01:06:57] particular warehouse application and we
[01:06:59] are uh enabling uh that for uh for other
[01:07:03] kind of platforms as well. Now this is
[01:07:06] where Lumotive is very much like a a you
[01:07:10] know semiconductor company as well.
[01:07:11] We're building the light control
[01:07:13] metasurface LCM chip which is an
[01:07:15] integrated uh SMOS chip enabling LAR
[01:07:19] platforms that really end up in total
[01:07:22] cost of ownership being much lower
[01:07:24] highly scalable for the robotic and
[01:07:26] autonomous platforms enabling much
[01:07:29] higher data quality especially when it
[01:07:30] comes to artifacts like saturation um
[01:07:33] like multipath and uh and outdoor
[01:07:36] performance even for IT sensors by the
[01:07:38] way um and then adaptive sensing
[01:07:41] enabling ing you to do different uh kind
[01:07:42] of performance uh in different regions
[01:07:44] of interest and making it highly dynamic
[01:07:46] for changing and very dynamic
[01:07:48] environments. Right now the way we go to
[01:07:50] market is uh we have the LCM technology
[01:07:53] and we have multiple partners across the
[01:07:56] industry including people like Namuga
[01:07:58] who are in South Korea and uh really
[01:08:01] focused on the outdoor uh robotic
[01:08:03] platforms. uh they're building sensor
[01:08:05] resistant technology with DTO uh you
[01:08:07] know DTO uh uh basically spat arrays out
[01:08:10] to 70 mters with 120 by 90 degree FOV
[01:08:14] and uh close to QBJ that's the best
[01:08:16] resolution today at least available
[01:08:18] until cogni comes up with something else
[01:08:20] uh in the spad based uh market we also
[01:08:23] have uh people like leopard imaging or
[01:08:25] partners like leopard imaging who are
[01:08:26] building RGBD sensors this is using an
[01:08:29] on semi sensor uh with one megapixel we
[01:08:33] use it in VGA mode 640 by 480 but with
[01:08:36] the LCM technology we're able to extend
[01:08:39] the range of IT from typically like 5
[01:08:42] mters in our uh architectures it's about
[01:08:45] 30 m so we can do 30 m of ranging with
[01:08:49] ITF with this type of resolution and
[01:08:52] then we are also developing several dev
[01:08:54] kits I put an example here which is a
[01:08:57] 180° sensor so full 180°ree by 110°ree
[01:09:01] using a single uh sensor Now this type
[01:09:03] of optical design can be applied to both
[01:09:05] DTO and ITO. We are not we're compatible
[01:09:08] across uh actually uh different image
[01:09:11] image sensors and our chip we have uh
[01:09:14] you know our gen 2 chip is for 9xx
[01:09:17] nanometer targeted for line scanning in
[01:09:19] LAR applications whereas we also have
[01:09:22] like a 1310 and 1550 nanometer chip. So
[01:09:25] we're not wavelength constrained right.
[01:09:27] uh we built that for the data center
[01:09:29] market but also we're getting a lot of
[01:09:30] interest from FMCW companies for the 13
[01:09:33] 10 15 nanometer chip to do solid state
[01:09:35] scanning for FMCW systems in the in the
[01:09:38] slow axis usually right but there are
[01:09:40] there are various things like you know
[01:09:43] possible so where our focus in the 3D
[01:09:45] sensing is really in the top because we
[01:09:47] think that the top can enable the lowest
[01:09:50] total cost of ownership in the system
[01:09:52] and can be scalable using the current
[01:09:54] ecosystem and enable us to do things
[01:09:56] like uh this highly dynamic sensing.
[01:09:58] Especially important for uh robotic
[01:10:01] applications where you know robots are
[01:10:03] in more dynamic uh you know use cases
[01:10:05] and you don't want to see you don't want
[01:10:07] to have static eyes you want eyes that
[01:10:09] can you know be task based and you can
[01:10:12] change how you see the world based on
[01:10:13] the task you're performing and we can al
[01:10:15] scale this as I said to full 180°ree
[01:10:17] sensing and we have these uh things
[01:10:20] coming out in the market through various
[01:10:21] partners. So just to sum it all up um we
[01:10:24] are uh you know taking a very much of a
[01:10:27] systems approach. By systems I don't
[01:10:28] mean just lighter or 3D sensing system.
[01:10:30] We mean the end platform system. So we
[01:10:33] are working closely with our partners to
[01:10:34] even integrate other sensors into the
[01:10:36] same package to really reduce the cost
[01:10:39] and make it easy to adopt right in the
[01:10:41] end platforms. We're extending the range
[01:10:43] of both ITO and TTO sensors to enable
[01:10:46] faster moving robots. We are enabling
[01:10:48] sensor consolidation. So one sensor
[01:10:50] instead of many saves a lot of dollars
[01:10:52] and complexity at a system level. We are
[01:10:54] solid state no moving parts in there uh
[01:10:57] ends up being reliable at least from
[01:10:58] shockw temperature range perspectives.
[01:11:01] Overall the power efficiency is much
[01:11:03] better at a system level both from low
[01:11:05] low power sensor or lightars but also
[01:11:07] less lightars or sensors overall in the
[01:11:10] in the system. We enable high high frame
[01:11:13] rates uh that is unprecedented in uh
[01:11:15] regions of interest. In regions of
[01:11:17] interest, we have shown one of our
[01:11:18] customers is using uh 450 Hz refresh
[01:11:21] rate uh in a narrow slice to look at
[01:11:24] really fast moving objects uh on the
[01:11:26] highway actually. Um high precision this
[01:11:29] is limited largely by the ITO and DTO
[01:11:32] but with steering and more power per
[01:11:34] pixel you improve the precision of uh of
[01:11:37] both of these technologies. So
[01:11:39] comparable uh architectures we have
[01:11:41] higher precision reduced noise and
[01:11:43] artifacts especially multiath and when
[01:11:45] you look at uh saturation from ground
[01:11:48] and uh you know uh shiny objects and
[01:11:51] software defined capability that really
[01:11:52] allow you to do closed loop sensing uh
[01:11:55] closed loop learning and optimization
[01:11:57] and one of the things that we enabled
[01:11:58] with a partner for this uh uh the heavy
[01:12:01] equipment is utilizing those
[01:12:04] capabilities to enable better learning
[01:12:06] models for edge AI application.
[01:12:08] So that partner and I'm I'm happy to
[01:12:10] actually Zach, I'm I'm interested in in
[01:12:14] testing out your dust chamber with their
[01:12:15] perception stack, not just with our
[01:12:17] sensors because they did develop a
[01:12:19] perception stack to see through dust in
[01:12:21] Saudi desert and they are deploying it
[01:12:24] there in heavy equipment from Mac and
[01:12:27] Mercedes and and such for their
[01:12:29] construction. So we were I was just
[01:12:30] there last week and we took a lot of
[01:12:32] data using a dump truck driving in the
[01:12:35] desert and and their dust uh kind of uh
[01:12:38] like you know cleanup algorithm or like
[01:12:40] removal algorithm worked beautifully
[01:12:42] right for the 360 application. So it'll
[01:12:44] be fun to connect and and see if we can
[01:12:46] bring some of that and test it out uh
[01:12:48] with you.
[01:12:49] Indeed.
[01:12:50] So that's all I had. I kept it short and
[01:12:52] sweet so we can keep uh some more room
[01:12:54] for questions and and discussions here.
[01:12:55] Thank thank you Apur. I mean Zah you
[01:12:57] already have a little uh I think uh
[01:12:59] group traveling to your to Fargo to your
[01:13:02] dust tunnel. So be prepared. Everyone is
[01:13:04] ready to test. Um I wanted to ask about
[01:13:08] so are you exact are you the similar way
[01:13:10] with the point cloud? Are you fabless or
[01:13:12] do you manufacture something yourself?
[01:13:15] We are completely fab. We are also using
[01:13:17] like leading semiconductor foundaries.
[01:13:19] We have multiple foundaries that we work
[01:13:20] with by the way because we want to make
[01:13:22] sure there's a multiple supply chain
[01:13:24] right diversity there. So you work
[01:13:26] multiple foundaries both in Asia and
[01:13:27] North America and same with the
[01:13:29] packaging houses right to package uh
[01:13:31] packager chips. So what we sell as a
[01:13:34] product is is these LCM chips that can
[01:13:36] be integrated but just like point cloud
[01:13:39] we do a lot of reference so our
[01:13:41] customers can go into building their own
[01:13:44] LAR or integrating directly because a
[01:13:46] lot of robotic companies or even you
[01:13:49] know automotive uh and and uh heavy
[01:13:52] equipment companies are becoming
[01:13:53] vertically integrated. they want to take
[01:13:55] build the sensors internally and
[01:13:56] integrate. So we are also working with
[01:13:59] several OEMs directly who can adopt a
[01:14:02] reference design and build their sensors
[01:14:04] and integrate it completely internally
[01:14:07] in their in their own kind of equipment.
[01:14:08] Right? So we have multiple go to market
[01:14:11] strategies
[01:14:12] uh but we do u uh you know enable full
[01:14:15] system designs as a starting point as a
[01:14:18] reference design. Mhm. So you can basic
[01:14:21] you you would be then to collaborate
[01:14:23] with the partners coordinate to develop
[01:14:25] the entire system together and yeah
[01:14:27] since you're apparently insensitive to
[01:14:30] the source or let's say not very agn
[01:14:33] that you're agnostic to the source much
[01:14:35] more possibilities open up and um I mean
[01:14:39] uh Zah had in mind a very long
[01:14:41] wavelength and maybe in this respect
[01:14:44] comes the questions do you any have do
[01:14:46] you really have any limitations for the
[01:14:49] wavelength that you can operate at
[01:14:51] because you mentioned 1.3 1.5
[01:14:54] what it's shorter or longer or it's
[01:14:57] doesn't matter. Yeah, for us like we are
[01:15:00] we have chips that are working um all
[01:15:03] the way from you know 900 nanometers
[01:15:05] actually down 80 nanome to 1550
[01:15:09] nanometers but we also have our views on
[01:15:12] the LAR market. So we do think that you
[01:15:14] know we're very excited about the recent
[01:15:16] developments in FFTW technology with
[01:15:19] some companies and I think Remis what
[01:15:20] they're doing is is really interesting.
[01:15:22] uh we also hear about van photonics and
[01:15:25] others right so we're keeping our pulse
[01:15:26] on the market uh and seeing where those
[01:15:28] are going but today if you just look at
[01:15:30] today what can be deployed at scale we
[01:15:33] think all of those are tobased sensors
[01:15:35] ITO and DTO because the only mature 3D
[01:15:38] sensing technologies outside of stereo
[01:15:40] which is very limited use cases for near
[01:15:42] range and and what you can do uh in and
[01:15:45] really 3D sensing what can go to market
[01:15:46] today and we are we need to make revenue
[01:15:48] today not three years from now right so
[01:15:51] from our reference design perspective
[01:15:53] and to make revenue today uh it it is
[01:15:56] all IPO and we think it is actually
[01:15:59] getting very interesting in direct of
[01:16:01] flight like with chips from on semi
[01:16:03] which are megapixel onchip processing
[01:16:05] and we extend the range to you know 30 m
[01:16:08] uh you know for robotic applications
[01:16:10] we're getting a lot of traction there
[01:16:11] because a lot of robots don't work in
[01:16:13] the dust like uh John Deere's uh you
[01:16:15] know heavy equipment does right so for
[01:16:18] for those type of like uh quadrupeds uh
[01:16:21] you know AGVA AMRS for even forklifts
[01:16:24] you know from people like uh Toyota and
[01:16:26] TMH and others we're getting a lot of
[01:16:28] interest for ITO sensors in time of
[01:16:30] flight sensors and that's uh actually a
[01:16:33] bigger uh like uh faster growing let's
[01:16:36] not say bigger but faster growing uh
[01:16:38] engagements for us are the IT and and
[01:16:41] kind of these 30 meter ITO sensors and
[01:16:43] then on the DTO side there is a lot of
[01:16:45] interest in the market but there's heavy
[01:16:47] competition from Chinese suppliers like
[01:16:49] Hessai and Robocense Right. So we see
[01:16:52] that for for the 3D for the deto market
[01:16:55] where we are getting traction is for
[01:16:57] people who don't want to use Chinese
[01:16:59] sensors right or one of these Chinese
[01:17:02] suppliers who want to extend the
[01:17:03] performance for the next it is it is but
[01:17:05] the rest of the supply chain is not
[01:17:06] there rest of the non-China supply chain
[01:17:08] is not there today it's not very mature
[01:17:11] so we are looking to see okay how does
[01:17:12] that market develop because as you said
[01:17:14] we are from Lumotive perspective we are
[01:17:16] very agile because we can support a lot
[01:17:18] of these different lighter architectures
[01:17:20] different wavelengths, different
[01:17:22] receivers, right? So, we are really
[01:17:24] seeing that okay, where is you know the
[01:17:27] need for wider field of view, for longer
[01:17:29] range, okay, and for uh more
[01:17:32] softwaredefined capabilities. That's
[01:17:34] where we can kind of really plug our uh
[01:17:36] our component in. Mhm. Yes. You
[01:17:39] mentioned uh uh in the presentation
[01:17:40] about software capabilities and I mean
[01:17:43] um how much because I remember I've seen
[01:17:45] one of your amazing demos and you also
[01:17:47] have this amazing scanning options when
[01:17:49] you can create different patterns and
[01:17:51] use them also for for scanning really
[01:17:53] with the patterning is it do you think
[01:17:56] and now you rather in the your
[01:17:58] presentation you concentrated on the
[01:17:59] software capabilities. So do you think
[01:18:01] it's uh let's say it's uh they are
[01:18:03] compatible or you rather decide okay
[01:18:05] that we don't we don't go anymore to the
[01:18:08] different or complicated patterns but
[01:18:10] rather put more effort into the signal
[01:18:12] processing and the software um let's say
[01:18:16] enhancements. M
[01:18:17] yeah actually for us it goes hand in
[01:18:19] hand Olga right it is actually the the
[01:18:21] software of perception software is
[01:18:25] becoming better because of the low-level
[01:18:28] programmability offer for example like
[01:18:31] you know what software programmable beam
[01:18:33] sharing does is where I want to get more
[01:18:35] precision and what r more range I can
[01:18:37] focus I can make the beam smaller so we
[01:18:39] don't just do steering by the way I
[01:18:41] think from our presentations to make it
[01:18:43] better we also change the we also have a
[01:18:45] lens function so we make the beam narrow
[01:18:47] or wider with our LCM while we are
[01:18:49] steering. So let's say in the horizon
[01:18:51] you need to look farther or with more
[01:18:53] precision or longer. We make the beam
[01:18:56] smaller so there's more power per pixel.
[01:18:58] Okay. And we can integrate longer and
[01:19:01] now you get better data to begin with
[01:19:03] and the perception algorithms run on
[01:19:05] better data producing better results.
[01:19:07] And when we're looking at the floor by
[01:19:09] the way we can make the beam bigger. We
[01:19:11] can expand the beam with the LCM using
[01:19:13] the uh the patterning on the LCM and we
[01:19:15] can you know reduce the amount of time
[01:19:17] we spend there because the ground is
[01:19:19] very close to these robotic systems.
[01:19:20] It's not very far right. So we can
[01:19:23] control the exposure. We can control the
[01:19:25] number of rows we are eliminating and uh
[01:19:28] you know the time we are looking or
[01:19:30] spending there. Now this is the part
[01:19:32] where it's completely flexible that the
[01:19:34] AI or perception layer can tune these
[01:19:37] parameters in real time. Okay. So as
[01:19:40] you're going let's say you're going
[01:19:42] through like you're driving in the
[01:19:43] driving mode they're like okay the speed
[01:19:44] of the vehicle is such and such I don't
[01:19:47] need to look at I only need to look at
[01:19:48] the horizon and get from point A to
[01:19:50] point B the perception stack will
[01:19:52] actually configure those sensors mode
[01:19:55] which is let's say the driving mode and
[01:19:57] when this vehicle or robot stops to do
[01:19:59] the work it can kind of change into a
[01:20:01] different mode to interact with the
[01:20:03] surroundings or the objects or get into
[01:20:05] a safety sense mode whatever the desire
[01:20:07] may be right so one of our humanoid
[01:20:10] companies loves this because as the
[01:20:12] humanoids are walking right those things
[01:20:14] they want to know where they're located
[01:20:15] but when they're at the table doing work
[01:20:17] they want to now see the objects in near
[01:20:20] range
[01:20:21] right so they're building a lot of this
[01:20:23] so this this actually makes us incumbent
[01:20:26] we sell we say okay we're solid state
[01:20:27] low cost all of that start trying us but
[01:20:30] when our customers perception team
[01:20:31] starts using all of these features
[01:20:33] that's where we become sticky then they
[01:20:35] can it's very difficult for them to now
[01:20:37] undo right uh or not use a lot of these
[01:20:40] features, they almost come to expect
[01:20:43] that, okay, this is what we need.
[01:20:46] Okay. Well, I hope in a few years your
[01:20:48] chip will be in one of my um uh vacuum
[01:20:51] robots. Meanwhile, that
[01:20:55] it's that it stops to to crush my lamps
[01:20:57] and other furniture. Uh meanwhile, we
[01:21:00] have a question from Ramos. Remos, you
[01:21:02] wanted to ask something the about the
[01:21:04] chip itself. Go ahead. Yeah, I mean it
[01:21:07] was just simple like uh was curious
[01:21:09] about
[01:21:11] loss budget for a single reflection and
[01:21:16] then um
[01:21:17] yeah absolutely yes so so for for the
[01:21:19] LCM technology right uh by the way all
[01:21:21] the numbers or all the claims that I'm
[01:21:23] making I'm making at a system level
[01:21:24] because we at the end of the day what we
[01:21:26] care about is when you enable a let's
[01:21:29] say a range a 50 m range what is the
[01:21:31] cost to the customer to get to a 50 m
[01:21:34] range and both in dollars and power
[01:21:37] right so it's really at the system level
[01:21:38] that the customers care about actually
[01:21:41] you know what like in the last year this
[01:21:44] is the first time I'm getting asked
[01:21:45] about the LCM efficiency itself because
[01:21:48] at this point most of the customers care
[01:21:49] about okay I'm getting a sensor for a
[01:21:51] range what is the power and cost of the
[01:21:53] full sensor solution that said our chips
[01:21:56] actually are about 50% efficient today
[01:21:58] so light in versus light out in the
[01:22:00] direction you want is about 50%
[01:22:02] we started breaking the cost and uh yeah
[01:22:05] we started breaking the cost and power
[01:22:07] budget compared to other illumination
[01:22:09] technologies for ITOH and DTO when we
[01:22:12] were at about 35% efficient. So, so now
[01:22:15] any improvement we make and we have a
[01:22:17] road map to get to about 65 to 70%
[01:22:19] efficiency. Any improvement we make
[01:22:21] basically lowers the cost or power
[01:22:23] consumption in the system further.
[01:22:27] How do I think about about uh angular
[01:22:31] range? Can I what's the max? What's the
[01:22:34] min and how much do I have? It depends
[01:22:38] on the angle.
[01:22:40] Let's continue the conversation maybe
[01:22:42] off the uh after we have always a little
[01:22:45] post seminar beer. Exactly. You can put
[01:22:48] it in the chat. Meanwhile, we are with
[01:22:50] the last two speakers with the these two
[01:22:52] speakers. We're already building up the
[01:22:53] system. So, we have a chip, we have a
[01:22:55] scanning system. Now we need a source
[01:22:57] and we have a a presenter in the room
[01:23:00] who will be responsible for the source
[01:23:02] part. Mitri tell us what do you have to
[01:23:05] offer for this comp for this novel um 3D
[01:23:09] sensing system that we're later going to
[01:23:11] offer to for Zach to test in his dust
[01:23:13] tunnel.
[01:23:14] Thanks a lot for the introduction. I
[01:23:16] think it's very very inspiring what we
[01:23:18] all looking at here. Um so share button
[01:23:22] I think you should see my screen right
[01:23:23] now.
[01:23:24] Yes. So the uh at TAS we make light
[01:23:28] sources. Uh so I think we all need light
[01:23:30] sources in a wild range of wavelengths
[01:23:33] and we started off in 2018 with a gluing
[01:23:36] simple chips together an indium
[01:23:39] phosphite chip and silicon nitrite
[01:23:40] feedback chip into these butterflies and
[01:23:43] then we had a a tunable laser. These
[01:23:46] tunable lasers can tune over 100
[01:23:48] nanometers tuning range and have a very
[01:23:50] narrow line width of 1 kohz. So that was
[01:23:53] very excellent for very small
[01:23:55] integrations via systems and and any
[01:23:58] other system there. In 2020 we had an
[01:24:01] evaluation electronics and in 2024 we we
[01:24:05] present nowadays the the light source.
[01:24:08] Here we step out the technology what we
[01:24:11] putting up to the market with we put uh
[01:24:13] product into the market but we can
[01:24:15] always go back to the uh uh to the chip
[01:24:19] tech if you want to co-integrate it into
[01:24:22] your uh yeah uh sensor system.
[01:24:26] Meanwhile we're expanding our wavelength
[01:24:29] portfolio. Uh we've got our workhorse at
[01:24:32] around telecom wavelength of the C band
[01:24:34] which use a nice 100 nanometers. uh and
[01:24:38] we also can go to slightly longer
[01:24:39] wavelengths. Uh this is interesting for
[01:24:42] fog applications in some uh cases. Uh
[01:24:45] for for dust, I don't know yet. Uh but
[01:24:48] yeah, perhaps we should find this out in
[01:24:50] the future. And uh the uh technology
[01:24:54] around 1310 is also emerging. So we've
[01:24:56] got a new product since one month on the
[01:24:59] market uh which delivers 100 nanometer
[01:25:02] tuning range around 1310 nanometers.
[01:25:06] got some specialties for some biomedical
[01:25:08] applications typically uh that's why I
[01:25:11] also interested in applications where
[01:25:13] you look spectroscopy uh spectroscopic
[01:25:16] features in the fields like raman or any
[01:25:20] absorption lines in the field. So there
[01:25:22] we also have some some light sources at
[01:25:24] shorter wavelengths here.
[01:25:27] Yeah. Where are the applications? Well,
[01:25:29] we see a ton of applications in this uh
[01:25:32] in this event that we're now looking at.
[01:25:35] But yeah, go lighter. Uh there's a lot
[01:25:38] of wind lighter applications in the
[01:25:40] field where you want coherence length of
[01:25:42] multiple meters. So you can say
[01:25:44] something about the uh the speed of the
[01:25:47] the the the air in in in the atmosphere.
[01:25:51] Uh there's photo acoustic imaging.
[01:25:54] There's an example of that. There's a
[01:25:55] laser pulse in penetrating into your
[01:25:58] into your skin or wherever it is. And
[01:26:00] then there is a a microphone array which
[01:26:03] is op yeah yeah read out by using optics
[01:26:06] and here our lasers find their way to
[01:26:08] the market to really do some 3D imaging
[01:26:11] inside your body uh for these kind of
[01:26:14] things. vibr vibration sensors uh lot of
[01:26:19] automotive applications run in this
[01:26:20] field uh for structural topology OCT
[01:26:24] where you really want to look into the
[01:26:26] depth of your tissue or material and
[01:26:30] then there's a lot of material science
[01:26:31] going on for fluoresence or anan or
[01:26:35] spectroscopy like like gas detection
[01:26:37] there
[01:26:39] um in spectroscopy and there's also some
[01:26:42] terraertz imaging going on some people
[01:26:45] want to use two lasers, beat them with
[01:26:47] each other and generate the terraheads
[01:26:49] uh image there to inspect welding
[01:26:52] systems and and material properties. So
[01:26:56] yes u we just make a tunable laser but
[01:27:00] the applications are are wild I must say
[01:27:03] and I think that's why it's very
[01:27:04] inspiring to to hear you all with the uh
[01:27:07] the applications in the field.
[01:27:10] This is the end of my presentation and
[01:27:12] it's open for any questions.
[01:27:15] Thank you Aditri. Then I will kick off a
[01:27:17] little conversation because you know uh
[01:27:19] as we're specifically in the 3D sensing
[01:27:21] meeting if you had a chance already part
[01:27:23] to work with the partners from the 3D
[01:27:25] sensing and what are their specific
[01:27:27] requirements that they will come to you
[01:27:29] and say does it become the coherence
[01:27:31] left down does it become the specific
[01:27:33] wavelength the power or other parameters
[01:27:36] that are like the most important to
[01:27:37] them. Zah I will later address this
[01:27:39] question to you for you to challenge the
[01:27:41] Dimitri can
[01:27:43] but Dimmitri first you if you have
[01:27:45] worked with 3D sensing guys what do they
[01:27:47] actually ask for
[01:27:49] and where can they sacrifice
[01:27:51] where yeah so that is always a challenge
[01:27:54] so thank you for the very interesting uh
[01:27:57] question so if you for example look at
[01:28:00] lighter systems our lasers are not very
[01:28:02] good in pulse mode so they need an
[01:28:04] external pulser which we don't have in
[01:28:06] in our systems typically
[01:28:08] Uh so we use FMCW lighter uh
[01:28:11] applications and the nice thing is when
[01:28:13] you have a small chir to do some some
[01:28:17] ranging uh the laser can tune to another
[01:28:20] channel and do the same trick there
[01:28:22] again. So with a single laser you can
[01:28:25] address multiple channels for example in
[01:28:27] the in the telecom band and if you have
[01:28:30] a very simple WDM network you can
[01:28:33] address multiple telescopes on your
[01:28:36] vessel drone or whatever you have there
[01:28:39] uh to look in multiple directions with a
[01:28:42] single light source there. So there we
[01:28:43] we we find our entrance there. Here for
[01:28:47] lighter especially the wind lighter you
[01:28:49] need long distances and there you need a
[01:28:52] very stable laser long uh coherence
[01:28:55] length and narrow line. There are
[01:28:57] external cavity is is is being exploited
[01:29:00] to the max and uh and it provides
[01:29:04] solutions that work with hundreds of
[01:29:06] meters distance there.
[01:29:08] Mhm. If you go for example with this
[01:29:11] very nice photo acoustic imaging
[01:29:15] well I've got a
[01:29:17] yeah a sketch over here is that uh while
[01:29:23] um yeah making some sound waves into the
[01:29:26] tissue there the sound waves propagate
[01:29:29] to the surface of your uh of your skin
[01:29:32] and yeah you need a microphone array to
[01:29:34] to digest everything there. So here
[01:29:37] there are multiple systems that that
[01:29:39] scan the region on vibrations. It's like
[01:29:42] a combination with fiber sensing to to
[01:29:44] to measure the deflection of the skin
[01:29:47] and to to to measure where the sound
[01:29:50] waves come from. So there you get a lot
[01:29:52] of uh yeah in-depth information there.
[01:29:55] Again here, fast sweeping or fast
[01:29:59] addressing channels is a key importance
[01:30:01] for for these guys because okay, the
[01:30:03] more pixels you can do, the faster you
[01:30:06] can build up your full image.
[01:30:08] Okay, Zach, how about you? So, would you
[01:30:11] like to ch uh to challenge Ditri and ask
[01:30:14] him for specific things maybe?
[01:30:17] Um
[01:30:19] hopefully not a challenge here but uh
[01:30:21] maybe Demetri can you talk maybe a
[01:30:24] little bit about some of the practical
[01:30:26] implications or issues you know going
[01:30:29] beyond 1550.
[01:30:32] Uh so yes our feedback tip is based on a
[01:30:37] silicon nitrite. Uh the techn technology
[01:30:40] can be made in in other material
[01:30:42] platforms as well but a new chip run is
[01:30:44] expensive hobby. Uh so uh at the moment
[01:30:48] we have the maximum wavelength on the
[01:30:49] shelf. It goes to close to 1,800 nmters.
[01:30:53] Uh and but there is actually we just
[01:30:57] need to rerun a chip in a different
[01:30:59] platform that that supports longer
[01:31:01] wavelengths. But I don't see a hard stop
[01:31:04] there. If you go for example to three,
[01:31:07] four, five micron wavelengths, I think
[01:31:09] that's easily feasible. uh but to to
[01:31:13] order a chip and design the chip that's
[01:31:15] expensive hobby but yeah I actually see
[01:31:19] that the longer the wavelength the
[01:31:21] easier the the platforms work they
[01:31:23] suffer less on on chip scatter events
[01:31:27] and and and other things where lasers
[01:31:29] are sensitive to so there I don't see a
[01:31:32] big issue
[01:31:35] and there are material thank you that
[01:31:37] support longer wavelength so I think
[01:31:38] that's a great asset
[01:31:41] Yeah.
[01:31:41] So that will Mitri your next guest to
[01:31:44] the ASA dust dust tunnel.
[01:31:48] Yeah, exactly.
[01:31:50] To see okay how far the 1700 penetrates.
[01:31:53] I have a question about 1550 because you
[01:31:55] know one of the selling points of
[01:31:56] luminina right was the 1550 uh for the
[01:31:59] eye um for the uh basically eye safety.
[01:32:03] But in order to get the usable power out
[01:32:05] of the system it had to be fiber laser
[01:32:07] which become very expensive. How is what
[01:32:11] is the story now?
[01:32:14] Yeah. So, I think very good question.
[01:32:16] People are always hungry for more
[01:32:18] photons and uh there's a good reason for
[01:32:20] that. Our lasers are diet lasers. Uh and
[01:32:24] out of these lasers, you get some 40 m
[01:32:27] of output power maximum and then you
[01:32:29] really end up with an external amplifier
[01:32:32] to boost the the levels to to yeah
[01:32:36] somewhat level of output power. uh
[01:32:39] really multi- watt level that we don't
[01:32:41] have on the shelf but what level can be
[01:32:43] made fairly easily with with amplifier
[01:32:45] sections behind the laser so that's
[01:32:47] that's the way path forward
[01:32:50] okay thank you so much
[01:32:52] I have
[01:32:52] well then yes go ahead
[01:32:55] um so um I was kind of curious like what
[01:33:00] I mean actually typically is like
[01:33:01] there's a little bit of a misconception
[01:33:03] here we generally don't need anything
[01:33:05] better than like about 100 kHz Like but
[01:33:08] the the big issue is the cost.
[01:33:10] Yeah.
[01:33:10] What would be a cost point at which you
[01:33:12] can deliver a 100 kHz 20 mill is fine
[01:33:16] like power doesn't really matter that
[01:33:17] much as you pointed out because you're
[01:33:19] going to use another car anyway.
[01:33:21] So but the issue is
[01:33:24] we need to come down to like 10.
[01:33:27] Yes.
[01:33:27] Can you do that?
[01:33:28] Yeah. I think for one for a 100 kilohz
[01:33:31] uh libraries I think that there is there
[01:33:33] are some other solutions available which
[01:33:35] are not relying on hybrid integration
[01:33:38] here we really exploit the the low loss
[01:33:40] part of the silicon nitrite which is
[01:33:42] just a glass medium and the amplifier
[01:33:44] section
[01:33:46] in indium phosphides or any other
[01:33:48] material which makes the light there. Uh
[01:33:51] so this hybrid integration is then
[01:33:53] offering a narrower line width which
[01:33:55] might be overkill for your application
[01:33:56] what you see say there. So these lasers
[01:34:00] what you see over here one offset the
[01:34:02] price is about €10,000.
[01:34:05] Yeah,
[01:34:06] hard to use
[01:34:07] prices will go down. But if you want for
[01:34:10] 100 kHz line with I think there the more
[01:34:13] affordable solutions on the market. Uh
[01:34:16] direct emitters DFB lasers perhaps you
[01:34:18] compromise a little bit on tuning range
[01:34:20] but yeah I think there's there's a lot
[01:34:22] of alternatives available. Yeah.
[01:34:25] Thank you.
[01:34:25] So I think this conversation is really a
[01:34:28] polite answer to your question. will I
[01:34:30] will follow up with you and so and so we
[01:34:34] now had and the steering system and chip
[01:34:36] and the source now we need to integrate
[01:34:38] it all into the one system and with this
[01:34:40] we'll need the last speaker of this
[01:34:43] session will help us with Filipe from
[01:34:45] flex compute Filipe the attention of
[01:34:48] everyone is yours
[01:34:50] uh great and uh it's been a very
[01:34:53] interesting section so I'd like to share
[01:34:55] today uh some tools that flex compute uh
[01:34:58] offer that help the this industry in
[01:35:01] design and optimizing their their tools.
[01:35:03] So uh flex compute is a GPU native
[01:35:05] platform that will offer uh simulation
[01:35:08] solutions for the photonics and other
[01:35:10] industries. So I would say that
[01:35:11] following the analogy we offer tools
[01:35:13] that help companies to design the eyes
[01:35:15] of the sensors. So since we're a GPU
[01:35:18] native uh company, we can run the
[01:35:21] simulations very fast. So we bring down
[01:35:22] the time scale the simulations from uh
[01:35:25] medium to small size uh structures to
[01:35:28] seconds. So in one hour you can hand run
[01:35:32] hundreds of simulations of uh and could
[01:35:34] optimize some uh small structures but
[01:35:37] it's also very scalable. So we kind of
[01:35:39] unlock new possibilities for
[01:35:40] optimization. So we can simulate a very
[01:35:43] large metal 10 billion grid points in
[01:35:45] the time scale of 1 hour. Uh that could
[01:35:48] not be done a few years ago. So you can
[01:35:50] have a full meta surface for bean
[01:35:52] steering for example and you can
[01:35:53] optimize the uh whole structure don't
[01:35:56] need to rely more in uh single unit
[01:35:59] cells and semical models uh and also
[01:36:03] offers some very powerful tools for uh
[01:36:06] optimization. One of them is inverse
[01:36:08] design. So the idea of inverse design is
[01:36:10] that you can uh have the gradients of a
[01:36:13] of a model for a given number of
[01:36:16] parameters. So this is the extreme case
[01:36:18] that each uh pixel is a free variable
[01:36:21] and can optimize to
[01:36:24] uh produce our logo in the fire field.
[01:36:28] Uh and that together with the
[01:36:29] scalability we can have a arbit number
[01:36:32] of parameters and quick optimize uh a
[01:36:35] sensor or a component for a sensor and
[01:36:38] you have everything integrated with
[01:36:39] different physics. So we can go uh in
[01:36:41] the same workflow uh simulate the optic
[01:36:45] RF heat charge. So this is an example of
[01:36:48] a avalanche photo diode that you can
[01:36:50] have everything in the same workflow. So
[01:36:51] you don't need to go back and forth with
[01:36:53] different softwares transferring data.
[01:36:55] It's everything the same uh the same
[01:36:57] workflow.
[01:36:59] Uh but the eye uh is not enough. So
[01:37:03] actually to see we need the eye and the
[01:37:05] brain. So in this analogy we don't need
[01:37:08] uh we actually need not only the single
[01:37:10] component but also a full uh photon
[01:37:13] integrated circuit. And for this task we
[01:37:16] offer this tool what we call photon
[01:37:17] forge is a photonic design automation
[01:37:19] tool that we call uh the main advantage
[01:37:22] that is integrated with a given foundry
[01:37:24] pdk. So you can uh build your layout uh
[01:37:27] use the the PDK uh parametric cells. You
[01:37:30] can build your own parametric cells. You
[01:37:32] have auto routing auto detect the parts.
[01:37:34] You can automatically connect components
[01:37:36] and it's very powerful because once you
[01:37:38] have uh your layout, you can assign uh
[01:37:42] models for each one of the components
[01:37:43] and carry out circuit level simulation.
[01:37:45] So if you need a full wave simulation of
[01:37:47] a given component, we call we
[01:37:49] automatically build a FDD simulation
[01:37:51] using our solvers solve that uh for each
[01:37:55] component and carry out circuit level
[01:37:56] simulation. So you can characterize uh
[01:37:59] with very accuracy uh a very large and
[01:38:02] complex photonic integrated circuit like
[01:38:04] this slider. And I'd like to talk about
[01:38:06] a little bit of what we have been
[01:38:07] testing today. It's a a gentic photonic
[01:38:09] design. Uh a very interesting feature of
[01:38:13] T3D is that we are Python native. So we
[01:38:15] can control all of our solvers with a
[01:38:17] very well doumented Python API and we
[01:38:19] also have very extensive learning
[01:38:21] resource u hundreds of examples uh a lot
[01:38:24] of documentation and that's gold for the
[01:38:26] LLMs. So what I have tried now is uh
[01:38:29] we've placed together an MCP that guides
[01:38:31] a LLM through our documentation through
[01:38:33] our API. So it knows how to use Tit 3D.
[01:38:36] I have plenty of information many
[01:38:37] different photonic designs uh and can
[01:38:40] now just ask for LLM to build a Tit 3D
[01:38:43] model but we are uh going further. We
[01:38:46] are actually trying autonomous design.
[01:38:48] So ask the LLM to start with initial
[01:38:51] design in this example this Ysplitter
[01:38:52] and autonomously trying uh engineering
[01:38:54] decisions to optimize that and it goes
[01:38:57] autonomously trying uh different designs
[01:39:00] and to optimize the full the final
[01:39:02] design and this design is pretty good.
[01:39:05] It reach more than 99% transmittance uh
[01:39:08] and was done about two hours
[01:39:10] autonomously and this is very flexible.
[01:39:12] It can be used for different type of
[01:39:14] problems. We have tried that for auto
[01:39:16] routing in the in aonic chip for uh
[01:39:22] for design P and junctions and so on. So
[01:39:24] it's very promising tool that have been
[01:39:26] testing with great success uh with
[01:39:28] different uh different type of problems.
[01:39:31] Well that's uh what I what I have to
[01:39:34] present today. So any questions and
[01:39:36] comments are welcome. Thank you.
[01:39:39] Thank you very much for being in the
[01:39:40] room with us today. Flex compute you
[01:39:43] have been supporting optica in many ways
[01:39:45] but today you are really bringing a very
[01:39:47] interesting topic which is which is the
[01:39:49] the use of metal lenses in lighter and
[01:39:51] when you saw before the the LCM
[01:39:53] structure presented by Lumotive so the
[01:39:55] question is very clear how can you help
[01:39:57] them they also can extend this to even
[01:40:00] 190 degrees beam steering is there a way
[01:40:02] that you could help the
[01:40:04] this is very interesting I think our
[01:40:05] team is already working with Tidy3D in
[01:40:07] in some ways so
[01:40:11] so The question goes back to you Aurora,
[01:40:13] how can you challenge them? What do you
[01:40:14] need from them?
[01:40:16] Yeah, I think I think uh you know from
[01:40:18] uh you know from from our semiconductor
[01:40:20] side, you know, we we are very impressed
[01:40:22] with the capabilities that Tidy3D is
[01:40:24] bringing. It's not uh easy to simulate
[01:40:26] so many different pixels and uh you know
[01:40:28] and and over the entire surfaces. So
[01:40:31] honestly, I personally don't know the
[01:40:32] details, but I know that my team is
[01:40:34] pretty happy with what they're seeing
[01:40:35] from the simulation environments that
[01:40:37] they're producing. So we're looking
[01:40:39] forward to more collaboration.
[01:40:42] I'm looking forward to more
[01:40:43] collaboration with you Philip from Fles
[01:40:45] Compute
[01:40:47] Pulva. You have been great today at the
[01:40:49] meeting. I would like to also say that
[01:40:50] my very friend very good friend Remos
[01:40:52] has been great and Hrik, but I would
[01:40:54] like to go to Sak once again for the
[01:40:57] closing remark. Sack, you have heard
[01:40:59] today about point cloud. You have heard
[01:41:01] today about the LCM of flex of of
[01:41:04] lumotive. You have heard about the
[01:41:06] simulation of flex compute. What is what
[01:41:08] is this idea that you go now with?
[01:41:13] Yeah. Um I think one of the things I
[01:41:16] walk away from, you know, learning is
[01:41:19] just really about the whole ecosystem,
[01:41:22] right? Uh that goes into developing
[01:41:26] um you know, especially these these LAR
[01:41:28] devices and how the whole ecosystem is
[01:41:31] becoming
[01:41:32] uh more mature, right, and evolving.
[01:41:36] uh you know it's not something that I
[01:41:39] follow that deeply right on a on a
[01:41:42] day-to-day basis. So it's fun to come
[01:41:44] into a forum like this and do a deeper
[01:41:48] dive and uh understand how it's evolving
[01:41:52] and how uh you know just gives me an
[01:41:55] impression that um you know these this
[01:41:59] technology
[01:42:01] uh is going to have a very important
[01:42:03] role to play right in our equipment
[01:42:05] moving forward.
[01:42:07] and uh that maybe we're we're just
[01:42:10] really kind of starting to scratch the
[01:42:11] surfaces of the of the capability and
[01:42:14] excited to see how it's going to
[01:42:16] continue to evolve.
[01:42:18] Let's make a deal. If you need the
[01:42:21] lasers from Chile, from Dimmitri, or you
[01:42:23] need the the development kit of Remos, I
[01:42:26] pay for the shipping expenses and I
[01:42:28] don't even charge it to keep in my own
[01:42:30] pocket. Thank you very much everyone for
[01:42:33] a fantastic fantastic meeting.
[01:42:42] This is optical.
[01:42:52] This is lighting red.