- il y a 2 jours
AI, Robotics, Humans and The End of Work As We Know It?
Catégorie
🤖
TechnologieTranscription
00:14Sous-titrage Société Radio-Canada
00:58Daniel Diaz
01:00There he is, with Digit.
01:02So, I'm from Agility Robotics.
01:04We create humanoid robots, and they are robots with two arms, two legs,
01:07and they're meant to do work, manual labor.
01:11And so, is this the end of labor as we know it?
01:13I think we're at the precipice of a fundamental change
01:17in the way we think about labor
01:19and how people are used or deployed to do work, absolutely.
01:25Across the globe, United States, in Europe, in Asia, any developed economy,
01:30what we're seeing is a massive shortage in the ability to get human labor
01:35into factories and warehouses.
01:37In the U.S., it's over a million unfilled positions.
01:40They cannot find people to take these jobs.
01:42And it's causing a huge issue in productivity.
01:46And as we start to talk about things like manufacturing and develop economies,
01:51it's really impossible unless you find solutions like Digit
01:54to step in and do these manual tasks.
01:58You know, it's a tool.
02:00Robots like Digit are a tool to help companies and people
02:04do work more efficiently, faster,
02:07and really enable people to do and focus on the things
02:10that people really should be doing.
02:12Higher order thinking, creative thinking,
02:15things that human beings are really just unique
02:17in their ability to accomplish.
02:18And, you know, it's interesting.
02:20This is so timely to talk about this right now
02:22because it feels like there are issues like robotics and AI
02:26that we have, of course, talked about for decades.
02:27We've imagined in science fiction for 100 years or more.
02:32But really, right now, we're experiencing a moment of transformation
02:36that is unlike anything I've experienced in my life,
02:39even having lived through the Internet, the mobile phone, broadband, etc.
02:42And now there is, we're really confronting these questions
02:47about will robots replace us?
02:49Will AI replace us?
02:51And the vision that you're sort of painting there
02:52is it's not exactly augmentation,
02:55but it's a sort of expansion or a shift in the label market.
03:00That's right.
03:00I think technologies like digit are absolutely going to fundamentally change
03:05how we think about work and labor.
03:07But this is on the scale of the iPhone.
03:11This is on the scale of electricity
03:13when you think about what sort of massive change is going to happen.
03:16So I don't want to underplay how radical a shift robots like digit
03:20are going to bring into the economies in the world.
03:25But every major shift, every big leap in technology has caused a change in the workforce,
03:33in the landscape of work.
03:34But it hasn't decreased the number of people who are in the workforce.
03:37There are definitely changes that we have to contend with,
03:40but we didn't see mass unemployment after the internet sort of fully took hold.
03:45These are tools.
03:46They're tools that people can use to do their work and do them better.
03:49No one would question a farmer choosing to use a tractor
03:53as opposed to hand plowing the field.
03:56This is going to be another one of those moments where 20 years from now
03:59we're going to look back and we'll say,
04:00of course we're going to deploy these technologies to do this
04:02because people are going to be doing very different types of jobs
04:05or doing their jobs in a very different way.
04:07Yeah.
04:07And so, you know, we've all lived in that era where we're online.
04:12We see these videos.
04:14There's so much hype now about robotics.
04:17It's hard for the average person to really comprehend
04:20how meaningful some of that is.
04:22But what is your sense about the hype versus reality
04:24when you see all this stuff going viral on Twitter?
04:29Right.
04:30I think there is a lot of hype in this market.
04:33I think there's a lot of implied functionality
04:37in a lot of the content that you're seeing online.
04:39There are, you know, a lot of big personalities
04:41at least, you know, leading these humanoid robotics companies.
04:44And I think they're pushing the boundaries
04:47of what's actually possible right now.
04:49It's great for looking at the future
04:52and thinking about a vision of, you know,
04:54how will we deploy these robots?
04:56Eventually, will they be in the home?
04:58How will that work?
04:59But the truth of the matter is,
05:01I don't think a lot of these companies have robots
05:03that are fully functioning,
05:04that are capable of being deployed and utilized commercially.
05:08A lot of the competitive set is really banking on
05:12full-stack control using AI
05:14and being able to deploy their robots in that way.
05:18And today, the data simply doesn't exist
05:21in order for companies to do that.
05:22It's the reason why you don't see
05:25all the humanoid companies deploying robots
05:27in factories and in warehouses.
05:30It's the reason you don't see
05:32other humanoid companies on stages like this.
05:34Digit has been quite literally around the world,
05:37in addition to being on the factory floors
05:40at Scheffler, the German auto parts manufacturer,
05:43at GXO, a big logistics player in the U.S.,
05:45in Amazon facilities, doing real work.
05:48Not in a lab, not in an R&D environment,
05:51but on the floor doing work.
05:53You know, agility gets paid for those robots
05:54doing that work.
05:55We are singular in that ability
05:57to commercially deploy these robots right now.
05:59And I think it's because we've taken
06:00a very unique approach
06:02to combining a traditional robotics control stack
06:05and layering that on with AI
06:07where the AI is most effective right now.
06:10And that's really in skills development
06:11and training those robots.
06:13We were talking earlier about, you know,
06:15you were reminding me that, of course,
06:17there already are millions of robots
06:19deployed on certain types.
06:21Not necessarily ones that look like humanoid,
06:23but we've been,
06:24the automation really is a spectrum
06:26rather than just like it suddenly popped up
06:29a week ago or two years ago.
06:31So if I look at Digit and I look at Agility,
06:34what's stopping you from having
06:3610 million of these deployed today?
06:39Sure.
06:41I think the biggest barrier
06:42to scaling robots like Digit
06:44is safety, for sure.
06:47There is currently,
06:49when you deploy a humanoid robot like Digit,
06:51it works within a confined area
06:54that is separated from human beings.
06:56It's in what's called a safety cage.
06:57And that could be,
06:58it could simply be a number of conveyor belts
07:00that are surrounding the robots,
07:01but it's barriers so that human beings
07:03can't get close to the robot.
07:04And the reason for that
07:06is that currently no humanoid robot
07:08is able to predictably,
07:11safely detect human beings
07:13in its environment
07:14and then react appropriately
07:16to ensure that the robot
07:17is not going to harm the person.
07:19How do you ensure that the robot
07:20doesn't fall on a person?
07:22How do you ensure that the robot
07:23doesn't accidentally move
07:24and hit a human being
07:28or if it gets bumped
07:29by another robotics platform
07:31like an AMR, a cart
07:33or a forklift,
07:34how do you ensure that robot
07:36is not going to fall
07:36in the direction of the human?
07:38That is called cooperative safety.
07:41That level of safety
07:41where the robot and the human
07:43can be side by side
07:44and you can guarantee
07:45that the robot won't harm the human
07:47is the biggest barrier
07:48to scaling this technology
07:50because you simply can't scale
07:52a hundred robots
07:52if every one of them
07:53is surrounded by these cages
07:55or these barriers.
07:57We've been...
07:58Number one priority for agility
08:00is ensuring human safety,
08:01safe operation of these robots.
08:03We've been working very hard
08:04at creating the world's
08:05first cooperatively safe robot.
08:08We have two prototypes
08:10that will be delivered this year
08:11and we will commercially deploy
08:13in 2026
08:14the world's first cooperatively safe robot
08:16that can be put to work
08:18side by side with people
08:19and guarantee it safety.
08:21That will be the first robot
08:23to actually open up this market
08:24and that's when you're going
08:25to start to see thousands,
08:27tens of thousands
08:28and eventually,
08:28your question,
08:29millions of robots in the world.
08:31And just to put that in context
08:32because people hear safety and robot,
08:34we're not talking about a robot
08:35going on a murderous rampage
08:37or becoming sentient
08:39and taking over the planet.
08:41We're talking about just the basic
08:43this thing knows I'm next to it.
08:45If that arm hits me in the face,
08:47it can break my cheekbone
08:48because it's a big heavy metal thing.
08:50So it's got to be able to know
08:51humans next to me,
08:53being aware,
08:54taking precautions.
08:55It has to be able to differentiate
08:56between static objects
08:57in its environment,
08:58machinery in its environment,
09:00moving robots,
09:01and a person.
09:02Because it can react differently
09:04depending on what's in its environment.
09:05It can be,
09:07it can act differently
09:08if it's a person next to it
09:09or if it's a column of a building.
09:11And so that,
09:11the ability to understand
09:13its environment,
09:14what's in it,
09:14and how to react differently
09:15around different objects or people
09:17is the critical factor there.
09:19It's a very difficult problem to solve.
09:20This is not simply a matter
09:22of having a camera on the robot.
09:23It is a full integration of sensors,
09:26it's software,
09:28it's the physical capabilities of the robot
09:30to react to the environment.
09:32You know,
09:32there's some very powerful motors
09:34in the hips and knees of this robot
09:35that enable it to perform tasks
09:38that other robots simply can't.
09:39and if you look at the competitive set,
09:42I think a number of them
09:43are going to have to think about
09:44some ground-up engineering,
09:45especially around the lower half,
09:47in order to accommodate
09:48what we know will be
09:50the safety standards coming out
09:51for humanoid robots.
09:53Agility is one of the leaders
09:55in defining the safety standards globally
09:58for how these robots will need to be built
10:01and what safety regulations
10:03they'll have to meet.
10:04We're helping drive that
10:05along with government bodies
10:06and other companies
10:07in the private sector.
10:09and we are sure
10:12that we're going to meet
10:13or exceed those standards
10:14because safety is paramount.
10:16I think the rest of the market
10:18is going to have to do some catching up
10:19on making sure
10:20that their robots
10:21will be able to be deployed
10:22in a safe way.
10:23And very quickly,
10:24you touched on this a bit before,
10:25but in terms of the business model here,
10:27is it robot as a service,
10:30is it SaaS,
10:31is it owning them outright?
10:32What does that look like?
10:33Probably, like,
10:34it's a good idea
10:34that we should look at
10:36the use cases for Digit, right?
10:37So we have focused
10:39this company
10:39on a number of use cases
10:42in manufacturing and warehousing.
10:44There's a reason for that.
10:45There's the critical shortage of talent
10:46within those verticals,
10:47and there are very physical,
10:49highly repetitive tasks
10:50that they can't find people
10:52to fill
10:53because these jobs are boring.
10:55They cause a lot of injury.
10:57And as a result,
10:58there's a lot of absenteeism,
10:59there's a lot of turnover,
11:00and it costs companies
11:01a lot of money
11:02to continually have to hire
11:04and train these people.
11:05We have a number of customers
11:07who actually overstaff
11:09every shift by 20%
11:11because they know
11:12they're going to have
11:13at least 20% absenteeism
11:14for these types of roles.
11:16Robots like Digit
11:17can easily fill that gap
11:19and really alleviate
11:20a lot of those issues.
11:22So we focus Digit
11:23on moving around,
11:24and I think there's a video
11:25we can show
11:25of the use cases
11:26that we have Digit,
11:27these plastic totes.
11:28In factories and warehouses,
11:30you will have
11:31highly automated solutions
11:33where these things
11:33are all moving autonomously
11:35through the factories
11:36on conveyor belts
11:37or on different robotic carts.
11:39But inevitably,
11:42one automation system
11:43will meet another one.
11:44So a robot cart
11:46loaded with totes
11:47will pull up
11:47to a conveyor belt,
11:48and a human being
11:50still has to move
11:51that tote off the cart
11:52and onto the conveyor,
11:53and that's somebody's job
11:54all day long.
11:55And there are a million
11:55examples of this.
11:56So those two points
11:58where the automations meet
11:59is where Digit can step in
12:00and fill those labor gaps.
12:01How do we charge for it?
12:03How do people procure
12:04these robots?
12:05There are two ways to do it.
12:06You can simply buy the robot,
12:08and we deliver ROI.
12:11If you buy the robot,
12:12the ROI on how much money
12:14you save by putting a robot
12:16to work instead of having
12:17a human being perform
12:18that task takes about
12:201.9 years.
12:21The other model of doing it,
12:23which is the most popular
12:24at this time,
12:25is the robot as a service model.
12:26The robot as a service model
12:27is where we take a look
12:29at your operation.
12:30We see how much
12:31you're paying for
12:32the equivalent of human labor,
12:34which they can't hire
12:35in the first place.
12:36And then we simply discount
12:39their fully burdened
12:40human labor rate.
12:41So the minute the robot
12:43steps onto the job,
12:44it's already doing it
12:45for cheaper than the human being.
12:47And so they're receiving ROI
12:49on day one
12:50of that robot working.
12:52Right now,
12:53we're focused on those areas
12:55where you can't find people
12:56to take these jobs.
12:57and so I think
12:58we are going to continue
12:59to focus on those areas
13:00where we're filling
13:01a labor gap
13:03and not this idea
13:04of we're taking jobs.
13:06We're going to fill those gaps
13:07that they simply
13:08can't find people for.
13:10And there is a very,
13:10very large company
13:12that we could build
13:13just on focusing
13:14on that right now.
13:15Okay, so speaking
13:17of use cases,
13:18let's see how this thing works.
13:20You want to see how it works?
13:20Yeah, can you take us
13:21to a demo?
13:22So let's talk a little bit
13:24about what Digit does
13:24and how it does it.
13:25So today,
13:26what we're going to show you,
13:27this is Digit, by the way.
13:29This is our humanoid robot.
13:31This design,
13:32also very intentional.
13:34So a lot of,
13:34you know,
13:34there's high Digit.
13:36And so these grippers
13:38are specific
13:39to the task at hand today, right?
13:41And so we have spent
13:42a lot of time
13:43really debating
13:44whether or not
13:45we think
13:45a five-fingered hand
13:46or specific grippers
13:48are the right way to go.
13:49Right now,
13:50we don't see
13:52a ton of customers
13:53asking for use cases
13:54that require
13:55a five-fingered hand.
13:56And so we have one.
13:58Digit can use five fingers
13:59and it has a hand
14:00that we can put on
14:00that does that.
14:02But so many of our customers
14:03have developed
14:04custom grippers
14:05that for the task
14:06that they're performing,
14:07we simply can put
14:08those onto Digit.
14:09That is the best way
14:10for us to do it.
14:12So today,
14:12what we're going to do
14:13is we're going to show you
14:14a demo.
14:15We layer a lot of AI
14:17into what we're doing
14:18with Digit.
14:18And today,
14:19what we're going to show
14:19is we're going to use
14:20an LLM
14:22to give Digit some commands.
14:23So I think today,
14:24guys,
14:24we're using Gemini
14:25is what we're doing?
14:26Yeah.
14:27So we're using Gemini
14:28and we're going to
14:29give Digit some tasks.
14:30What we've done
14:31is we've put some laundry
14:32in front of Digit.
14:33We've got a little basket there.
14:35One of the things
14:35that I imagine Digit doing
14:36for me in a number of years
14:38when it can actually
14:38enter in my home
14:39is putting away my laundry
14:40because somehow
14:41the laundry gets done
14:42and it ends up in piles
14:43in the kids' room
14:44and nobody ever puts it away.
14:45So I would very much
14:46like to see Digit tackle this.
14:47But we can ask Digit
14:48to do a number of things.
14:49So we've got
14:49different colored T-shirts
14:51here on the table.
14:52We can just start to ask Digit
14:54to start to sort this laundry for us.
14:56And can we ask it
14:57to sort and fold
14:58or just sort in?
14:59Today, we're just going to sort.
15:00Okay.
15:00So if we ask
15:01to put the red shirts
15:02in the basket?
15:02Sure.
15:03So both or just one?
15:06Let's do both.
15:07Okay.
15:08And by the way,
15:09as you're doing that,
15:09I believe on the screen
15:10people will be able to see
15:13how they're processing the thought.
15:15That's right.
15:15So what you'll see
15:16through the LLM
15:17is you're going to see
15:18Digit processing my command
15:19and what it's going through
15:20as it starts to think
15:21about what it's doing
15:22and then how it's going
15:22to perform its task.
15:24So let me give Digit a command.
15:27Pick up the red shirts
15:28and put them in the basket.
15:36All right.
15:41So here Digit is saying,
15:42all right?
15:47And so while he's doing that,
15:49or he, I'm calling it he,
15:51it, take me through
15:54how it's seeing that.
15:55What are the components?
15:57What is the hardware
15:58that's allowing it
15:59to sort of detect these things?
16:01Right.
16:01So Digit has a number
16:02of RGB cameras,
16:03so I missed the basket
16:04a little bit there,
16:05but it's got a number
16:05of RGB cameras all around,
16:07and it's got LIDAR here,
16:08and so what it's doing
16:09is it's seeing
16:10what's in front of it.
16:11It's determining
16:12which ones are red
16:13and it knows exactly
16:14where they are.
16:15It's going to pick them up
16:16and then it's going to go,
16:16it knows that's a basket
16:18and it's going to,
16:18it's going to drop them in there.
16:20So it's doing all of this.
16:21This is not programmed.
16:23We don't have,
16:23I mean, I asked you
16:24which color you wanted
16:25the shirt to be.
16:25So we can ask it
16:27to do any number of things.
16:29It's not a pre-programmed demo.
16:30So it is literally,
16:31you know,
16:31looking at the environment,
16:33interpreting the environment,
16:34and then carrying out the task.
16:36So I can ask it
16:37to do things like put the gray,
16:38there's a gray shirt
16:39and a green shirt on here.
16:40I can say put the gray shirt
16:41on top of the green shirt
16:43and then put the pile
16:44in the basket,
16:44something like that.
16:45So I can say that.
16:46So let's see.
16:50Put the gray shirt
16:51on top of the green shirt
16:53and put the pile
16:54in the basket.
17:01I'm going to try that again
17:02because it didn't hear me.
17:04It's the iPad
17:04that didn't hear me,
17:05not digit, sorry.
17:07put the gray shirt
17:08on top of the green shirt
17:10and put the pile
17:11in the basket.
17:16There we go.
17:24Now it's sending.
17:26So now it can figure out,
17:27it'll determine which one's gray,
17:28which one's green,
17:29and begin to order its operation.
17:31So there we go.
17:32Put the gray shirt
17:33on top of the green shirt.
17:36and now it can take them both
17:37and put that in the basket.
17:46I'll move the combined pile
17:48into the basket, it says.
17:50Now, oh, we have a miss.
17:55We try it again?
17:56Yeah.
17:57Digit.
18:02Digit.
18:03Let's try that again, Digit.
18:06Put the pile of shirts
18:07into the basket.
18:18There we go.
18:19You going to grab the red one?
18:20Nope.
18:21There you go.
18:22It's a little perception
18:23just there,
18:24but you get the gist.
18:25So we can start
18:26to use these LLMs
18:27to give commands now
18:28instead of having to actually
18:29program the robot.
18:30and so you think about
18:31a lot of the work
18:32we're doing now
18:32is utilizing AI
18:33to train these robots
18:35and so we can start
18:35to refine these things.
18:36Oh, thanks, Digit.
18:37And it's learning.
18:40So now I put the full stack
18:41in the basket.
18:42All right.
18:43That's great.
18:44That's Digit.
18:45So this is very,
18:47I mean, these are very basic tasks,
18:48but if you look up
18:49at what the robot is doing,
18:50it is like in factories
18:51and so right now
18:53we're doing all this
18:53bulk material handling
18:54where we're moving
18:55these big bins around,
18:56but the next step for Digit
18:58is to start sorting
18:58small materials,
18:59begin putting different orders
19:02together in these totes
19:03and then sending them along.
19:04And so really thinking
19:05about automating
19:06a lot of what happens
19:07in these factories
19:08and in these warehouses
19:09so that Digit can take on
19:11these highly repetitive
19:13physical jobs
19:13and people can start
19:14being deployed
19:15to do more interesting things,
19:17managing fleets of robots,
19:18managing the software,
19:19doing remote operations
19:20of these robots.
19:21There'll be a whole slew
19:22of digital jobs
19:23that'll be created
19:24once we begin to deploy
19:25these things in mass.
19:26And so again,
19:26it's a shift
19:27in the labor market,
19:29not a decrease
19:29in labor market.
19:30And very quickly,
19:31tell me again
19:31just about design
19:32because I've seen this,
19:34the knees,
19:35the legs on that robotics,
19:36that kind of ostrich
19:38or stork.
19:39Yeah, I'm not sure
19:40if everybody can see
19:40but Digit's legs
19:41are actually kind of bird-like.
19:43They go backwards
19:44and if we want to put up
19:45some more images
19:45of the robot.
19:47So we did that
19:48because it makes it
19:49a lot easier
19:50for Digit to actually
19:51get down and get close
19:52to what it's carrying
19:54and so we can actually
19:55get really close
19:55and bend down very,
19:56very low
19:57and it allows Digit
19:58also to be able
19:58to reach really,
19:59really high.
20:00The next iteration
20:00of the robot,
20:01the engineers have actually
20:02figured out a way
20:03to get the same level
20:04of force and power efficiency
20:06actually with the legs
20:07facing the other way
20:08and so you'll see
20:09our next version of Digit
20:11will look quite different
20:12actually.
20:13The next version of Digit
20:14will have a number
20:15of innovations.
20:15One, it's going to have
20:18that cooperative safety
20:19I was talking about.
20:20This robot,
20:20which comes out in 2026,
20:22will be able to work
20:22side by side with people
20:23but it also will have
20:24a world's first
20:25in a battery
20:27that has a 10 to 1 charge ratio.
20:29That's a battery
20:29that you charge for 10 minutes,
20:31the robot will work
20:32for 100 minutes.
20:33That's increased
20:34from a 4 to 1 charge ratio
20:36that we have now.
20:37So now you're looking
20:37at a robot
20:38that can work,
20:39one robot that can work
20:41three shifts a day.
20:42So very efficient
20:43for our customers
20:44and great for agility
20:46and its ability
20:46to drive profitability
20:47in the company.
20:48And the third thing
20:49that robot will have
20:50are swappable end effectors.
20:52so the robot will be able
20:53to by itself autonomously
20:55swap out the hands
20:56that it uses
20:57and be able to change them
20:58based on the job
20:59it's being asked to do.
21:00Those are the three things
21:01we think are big game changers
21:02for the whole category
21:04of humanoid robotics.
21:05So putting that in context,
21:06obviously,
21:07as people probably heard,
21:08there's a huge debate
21:09in the U.S.
21:10trying to reshore manufacturing,
21:12bring that back.
21:14You were talking before
21:14about for some of these tasks
21:16and we've seen this
21:17in a public debate,
21:18you know,
21:18if you could magically
21:20bring all these things back,
21:22would people want to do them?
21:24Would there be the people
21:24available to do them?
21:26Right.
21:27So,
21:28and then there's a third part,
21:30oh,
21:30they're going to bring it all back
21:31and just put robots
21:32in these factories.
21:33Sure.
21:33Who cares if they bring
21:34manufacturing back?
21:35So,
21:35where does
21:37Digits and robotics
21:38kind of fit into
21:39that conversation?
21:40I think Digit
21:40is a critical part
21:41of bringing manufacturing
21:43back to established economies
21:46for the reason
21:47that I mentioned, right?
21:47So,
21:48we can't find people
21:49to do these jobs now.
21:50The idea of bringing
21:51more factories back
21:52to the US
21:53or into Europe
21:54or building new factories
21:55in Asia,
21:56if we can't find the people
21:57for the factories
21:58that we have now,
21:59I'm not sure how we're supposed
22:00to scale that up
22:02if we think human labor
22:03is the answer.
22:04It's not going to be the answer.
22:05The answer is going to be
22:05automation platforms
22:06and technologies like Digit
22:08who are going to step in
22:09and do these jobs
22:09and in doing so,
22:12they will create
22:12a new level of job,
22:14a new type of job,
22:15digital jobs,
22:16where they simply
22:17don't exist, right?
22:18So, if we're building
22:19new factories
22:19and it's primarily
22:20driven by automation,
22:21there are still people
22:22who are going to have
22:23to manage,
22:24maintain those facilities.
22:25Those are net new jobs
22:27we're going to see.
22:27So, no,
22:28we're not going to bring back
22:29hyper manual
22:30repetitive labor
22:31to these markets.
22:33It's simply,
22:33there isn't a market
22:34of employees to fill that.
22:36But we will bring in
22:37new types of jobs,
22:38digital jobs
22:39that don't exist right now
22:40and I think that's
22:41where we're going.
22:41That's the shift
22:42we'll see in these labor markets.
22:44And on that sort of
22:45large macro
22:46competitiveness stage,
22:48there's also
22:49the issue of China
22:50which has been
22:51very aggressive
22:51about adopting these
22:53and let's just say
22:55their populace
22:57is a bit more pliant
22:57when it comes to
22:58not having too much
23:00of a say
23:00or pushback
23:01in these things.
23:02So, what do you see
23:03happening there
23:04compared to
23:04the US or Europe?
23:06I think we've seen
23:06some really impressive
23:08physical feats
23:09from the robots
23:09that are coming out of China.
23:10I don't think we've yet
23:11see them direct anything
23:13in a way that
23:13is really meant
23:15to derive value, right?
23:16or deliver value
23:17for our customers.
23:18So, it's cool
23:18that your robot
23:18can do parkour
23:19but I don't think
23:21anybody's paying
23:21for parkour robots
23:22in their factories.
23:24I think they have an issue
23:25with software development
23:27in general
23:28but the biggest factor
23:30that I think China
23:31will have to contend with
23:32when it comes to humanoids
23:33is trust.
23:34In order for these robots
23:35to operate in a factory,
23:37you have to upload
23:37digital twins
23:38of your factory,
23:39highly sensitive IP
23:41around your manufacturing processes.
23:43and there's just a lot
23:45of sensitive data
23:45and you have to entrust
23:47that to the robot manufacturer
23:48in order for these robots
23:49to operate
23:49in those facilities.
23:50And then,
23:51when you deploy this robot
23:52and you see Digit here
23:53has got a ton of cameras
23:55in its waist
23:56around its neck,
23:57it's got a LiDAR scanner.
23:58So, these robots
24:00are able to collect
24:01massive amounts of data.
24:04Do you want a walking camera
24:06in your facility
24:07filming everything you do
24:09if you don't trust
24:10the manufacturer?
24:11I think the biggest problem
24:12Chinese manufacturers
24:14of these robots
24:14are going to have
24:14is the level of trust
24:16required to deploy these
24:18in your facility.
24:19And I don't think
24:20they're going to get over
24:20that bar anytime soon.
24:21Outside of China.
24:22Outside of China.
24:23There will absolutely be markets
24:24in Asia
24:25and specifically in China
24:26where I think
24:27they'll be fine with that.
24:28But I don't see that
24:28happening in the US
24:30or here in Europe
24:31anytime soon.
24:32Yeah, I mean,
24:32if we're worried
24:32about TikTok.
24:34Seriously.
24:34If we're worried about TikTok,
24:35I don't think we're going
24:36to let these robots
24:36into our factory.
24:38Okay, a couple of fun questions
24:39that are going to wrap this up.
24:40Okay.
24:41As we all know,
24:43robotics have been
24:44such a part of our
24:45sci-fi culture
24:46and lore for so long.
24:48Yep.
24:49As you've seen
24:49the different iterations
24:50from, you know,
24:51the fun friend,
24:52the friendly robot
24:53and lost in space
24:54to murder bot
24:55and Terminator
24:56and all this stuff
24:56in between.
24:57Yep.
24:57Does that,
24:58what does that do
24:59to the calculation
25:00of the cultural acceptance
25:02and how do you see
25:03that as sort of helping
25:04or hurting the adoption?
25:06I mean,
25:07I think it starts
25:09to help people understand
25:10how eventually
25:11we'll see these robots
25:12become part of regular life
25:14and so that's helpful.
25:15I think how people
25:16want to interact
25:17with robots
25:18has been well established
25:19in these narrative mediums
25:20like movies
25:21and TV shows.
25:22I think what folks
25:24need to understand
25:25is where these robots
25:26are right now
25:27and what they're capable
25:28of doing.
25:29They're not murder bots.
25:31I think, you know,
25:32in the humanity category,
25:34agility specifically,
25:35safety is the number one
25:36thing we worry about
25:37and so we are deploying
25:38these very carefully
25:39and really only
25:40in environments
25:41where we can ensure
25:41humans are going
25:42to be kept safe
25:43and as they get more
25:44and more skilled
25:45and the safety protocols
25:46and the abilities
25:47of the robot
25:47get better and better,
25:48we'll deploy them
25:49in new,
25:50more diverse environments
25:51but I think it's a help.
25:53I think, you know,
25:53I think when people
25:55begin to understand
25:56and they have some context
25:57around how these robots
25:58will help us do new things,
26:00that's not a bad thing.
26:02Yeah, yeah.
26:02Okay, last question.
26:03If there was one myth
26:05you could bust
26:05about robotics right now,
26:08what would that be?
26:09One myth about robotics
26:11that I could bust right now.
26:14I think it's how quickly
26:16we will get to a robot
26:17that is fully controlled
26:19using AI.
26:20I think there's a lot
26:21of implied, you know,
26:23sort of videos out there
26:24and statements about,
26:25you know,
26:25it's going to be in a year
26:26or two years.
26:27I think that it's a longer
26:29way away than that.
26:30It's going to happen
26:31but I think it's not next year,
26:34it's not the year after.
26:35I think people need to understand
26:36that in order to do this
26:38we've got to make sure
26:39that we're covering
26:40on the safety protocols first,
26:41these robots have to be useful
26:44and AI will come
26:45as that data pool
26:46gets larger and larger.
26:47that data required for AI
26:50is a combination
26:51of a couple of different factors.
26:53You need a lot
26:54of simulation data,
26:55we've got a great partnership
26:56with NVIDIA,
26:57we train our robots
26:57in simulation
26:58and then there's data
27:00that comes from deploying
27:01the robots in the real world.
27:02If you can't scale
27:04your robots in large numbers,
27:06you'll never get
27:07that real world data.
27:08If you don't have
27:10a safe robot
27:11that can work outside
27:12of a safety cage,
27:13you're never going to get
27:14scaled real world data
27:15and that's why
27:16we've taken the approach
27:17we've taken.
27:17We combine simulation data
27:19and then verification
27:20through real world data.
27:22Those two things
27:23will drive the AI race forward.
27:26That's the thing
27:27that people have to realize
27:28that it's going to be
27:29that approach
27:30and it's going to take
27:31maybe a little longer
27:31than we'd like.
27:32And just to clarify
27:34something there,
27:34I referenced this before
27:35but we're talking
27:36about the AI component.
27:38That part of it
27:39is transforming faster
27:41than anything I've experienced.
27:42Just that,
27:43the software algorithmic
27:44part of it.
27:44anything I've experienced
27:46in 30 years.
27:47You guys are not developing
27:49your own LLMs.
27:50You're using other ones
27:52and that's a core debate.
27:54Anyone who's working
27:54in generative AI,
27:56agentic AI,
27:57do I build my own models
27:58in my vertical app?
27:59How do I think about that
28:00and how do I make sure
28:01that I don't bet
28:02on the wrong thing
28:04or bet on something
28:06that becomes the beta max
28:07to the VHS
28:08or whatever the analogy
28:09you want to use?
28:10We are agnostic
28:12to which LLM
28:13we're using.
28:14we're working with
28:15all the different players
28:15right now in AI
28:16because we're going to see.
28:18We'll see which one
28:19actually takes the lead
28:20in actually creating
28:20the foundation models
28:21that are best suited
28:22for embodied AI.
28:24So we're not placing
28:24a bet right now.
28:25We're also not trying
28:26to build our own.
28:27There are enough companies
28:28out there with expertise
28:29and with tons of funding
28:30who are going to do that.
28:31We're going to focus
28:32on the close collaborations
28:33to make sure
28:34that they understand
28:35that you cannot develop
28:37these things
28:38devoid of
28:39and clear integration
28:41with the actual hardware.
28:42So anyone trying
28:43to do it software only
28:44I think they're going
28:45to run into some problems
28:46that they haven't
28:48quite anticipated yet.
28:49And so it's going
28:50to be only through
28:51the close collaboration
28:52with hardware
28:52that you're really
28:53going to get to that point
28:54that we're all looking for.
28:55Yeah.
28:56And of course
28:56you're getting
28:57a specific set of data
28:58as well
28:59from the hardware
29:00from the way
29:01you've developed that
29:02that gives you
29:03kind of a
29:04I would assume
29:05an added competitive
29:07advantage there
29:08in terms of
29:09your moat
29:10your business.
29:11Real world application
29:12real world data
29:13is going to be
29:14a critical factor here.
29:16And so we're very
29:16excited about
29:18being the first company
29:19to scale this technology.
29:20Yeah.
29:20Well congratulations.
29:21Thank you very much.
29:22Everyone please
29:22a big round of applause
29:23for Daniel and Digit.
29:25Thank you.
29:26Thank you.
29:27Thanks Chris.
29:32Thank you.
Commentaires