【3万体のヒト型ロボットが人手不足を解決する】冷蔵庫も運べる「アトラス」が自動車を作る／元祖ロボ企業幹部が語る「30年の技術力」／人間の“コピペ”＋GeminiでAI開発が高速化【1on1 Tech】

[00:00:00]Wow.

[00:00:19]Dynamics.

[00:00:45]Okay, we have Brandon Schumman from Boston Dynamics here. Thank you for coming today.

[00:00:49]>> Thank you for having me.

[00:00:50]>> Yeah. Um, so Boston Dynamics has been at the very forefront of the robotics uh research and industry. Um I think it's one of the companies first companies to develop highly capable humanoid like Atlas. So and the robots that kept impressing the people online from videos on YouTube. But um I was wondering that what the company's uh core technological strengths are.

[00:01:19]So what's your take on that? I think we've been in the legged robot business for a long time, perhaps longer than anyone. Uh, starting, you know, over 30 years ago when we spun out of the Massachusetts Institute of Technology.

[00:01:35]>> Uh, when we pivoted to commercial applications about 7 years ago, >> we really have been focused on what the customer needs. So I I think there are a lot of companies that are very exciting that get into making the robots uh doing demonstrations. I love it. I think it's a very exciting time in the industry.

[00:01:56]>> Our strength and our focus is on delivering commercial value to customers in different industrial applications and I really think that that differentiates us not just by being first years ago but also now very focused on on satisfying customers.

[00:02:13]>> Mhm. is the the technological strength coming from the like a fundamental research at the MIT and the sort of the academic background.

[00:02:24]>> I think in part it's it's u having some of the most experienced roboticists in the world working for us but it's also relentless effort to make the technology reliable. If if you came to visit us in Boston Dynamics, you would see our spot robots walking around the building 24 hours a day.

[00:02:43]>> 24 hours a day.

[00:02:44]>> Oh yes. Over and over again. And this this has been happening for years.

[00:02:48]>> So the way we make spot very very reliable is we're just constantly testing it, waiting for it to fail so we can learn how to make it better.

[00:02:56]>> And we've been doing that for years.

[00:02:58]>> I I think that's that goes handinhand with with the technology development um capabilities. Mhm.

[00:03:06]>> Do you think the uh um the technological paradigm changed like after the like um the generative AI or LLMs are introduced in the world after like a emergence of chatbt?

[00:03:22]Um do you feel the acceleration of technology after that?

[00:03:26]>> Absolutely. There there's been a a huge I would call it a revolution in robotics. You know, we spent much of those 30 years deterministically programming robots like Spot to do to do what they do.

[00:03:40]>> But as of a few years ago, we really have pivoted to AI training. So if you if you look up our videos or our blog posts on on Atlas and how we are training Atlas to perform manufacturing type tasks. Yeah, >> it's all AI, right? So, it's it's reinforcement learning. It's what we call large behavioral models. So, in the in the Gen AI context, you have large language models.

[00:04:09]>> For us, it's not about language or pictures or text. It's about behavior.

[00:04:14]How do you get a robot to do something?

[00:04:16]You generate a large set of behavioral data and that then allows the robot to basically train itself to perform the task that you are showing the robot how to do. And that that really is quite different and has really fundamentally changed the way we develop these technologies.

[00:04:34]>> Um did that change how the company developed the robot? Like a fundamentally >> I think it it's changed how we program the robot >> to accomplish different behaviors. I think fundamentally you still need to be very good at hardware.

[00:04:53]>> Mhm. So AI is not going to build the robot for you, at least not yet. So you know, some of the innovations, for example, with Atlas is we're focused on having the legs be the same, so that in the event of of a repair situation, you can just use any leg. We also have designed Atlas to use the same two actuators across the entire robot.

[00:05:17]So we focused on making the hardware efficient and to lowering cost so we can deploy at scale commercially.

[00:05:24]>> That's the hardware side. You still need to be able to do that. Then once you have a machine that can physically walk and move >> and touch things, lift things, manipulate the world.

[00:05:35]>> Now you have to also work on the software side. How do you get that machine to do something useful and to do it reliably and repeatedly over and over again? So you need both to be successful.

[00:05:47]>> It's really a combination >> of hardware and software.

[00:05:50]>> Correct.

[00:05:50]>> Yeah. Um so you talk about um Spot and Atlas like Spot is the uh quadrop and the Atlas humanoids. So how do their roles differ from one another?

[00:06:04]>> When we designed spot the you know it was really in some ways an experiment.

[00:06:10]We had been working on legged robotics for a long time, starting at MIT, just to see if you could create a robot that could navigate the world and one day be useful at doing something.

[00:06:22]>> That something ended up being industrial inspection where we've seen the most customers for Spot has been factory inspection, automated inspection. So you have the robot basically doing preventative maintenance >> or sending data back to the control center to make sure the facility is running efficiently. We've also seen a lot of amazing uh results in public safety where spot is keeping people safe from harm.

[00:06:49]>> That could be a bomb disposal situation.

[00:06:51]It could be a police law enforcement >> uh situation with a dangerous suspect.

[00:06:56]So, I would say industrial inspection and public safety are the main uses for spot because of the form factor. It's low to the ground. It's good at scanning and acquiring data, but it is not going to replace a um a task where you need the ability to reach above your head or to do things that um a human form factor could. So, with Atlas, we we really are focused on sort of two things. number one creating a robot that simply can be used anywhere a person could be doing something but number two also to create a form factor that allows us to collect the data to train the robot using people. So using a like a visual VR um visor goggles and hand controllers looking through the eyes of the robot you can generate the data and because Atlas has a human form factor >> the human maneuvers to generate data can be transferred right to the robot. You couldn't do that with Spot because Spot has four legs and it doesn't have two arms. So there's a there's a distinction in how you develop the robot but also in how it's used. M >> and Boston Dynamics has started implementing uh Atlas like into the factories uh of Hyundai in the US I think and >> well well that that's the plan. So there'll be deployment starting later this year and and then ramping up next year.

[00:08:23]>> So we uh we are on the verge of deploying uh at Hyundai in Georgia.

[00:08:27]Correct.

[00:08:28]>> Yeah. And uh the company announced there's going to be around 30,000 robots by 2028. I think >> that's the plan. Yeah. And so what are they going to do specifically?

[00:08:39]>> Automotive assembly and manufacturing.

[00:08:42]Um so uh in in an automotive factory, you have pieces that have to be moved from um one sorting or or staging bin to another. So if you look at our videos online, you'll see examples of of Atlas doing that, moving things from one um one cart to another. Um and also um lifting things into place. really the the whole um manufacturing process that takes place in an automotive manufacturing cell could be all the kinds of things that we train Atlas to do. Once we have a model on Atlas that's capable of learning effectively, which is what we're developing, all of the tasks that could be performed, particularly lifting heavy things, can be done by the robot. And in the most recent example, we put out a a just a almost a fun demo of Atlas lifting a 45 kg refrigerator. So someone someone asked, you know, Atlas, can you bring me a drink? And instead of bringing just one drink as a joke, we brought the whole fridge. Um, but that shows you that this is a robot that can lift more weight than a human should in a factory setting. So we can make the manufacturing process more efficient.

[00:09:54]>> Yeah. Um how is uh the company preparing to implement those robots at such a scale?

[00:10:01]>> Well, we're ramping up manufacturing the the you know the plan is is to create a robot manufacturing facility in in the near future and to try to scale up from there and I I think in that respect Hyundai has been a terrific partner uh given that you know they create automobiles at scale. So we have a lot that we're learning from them in terms of making the robots at scale. Is it like the biggest uh deployment of robots so far for Boston Dynamics?

[00:10:29]>> Um oh for us? Yeah.

[00:10:30]>> Uh it it would be once once the deployment occurs it would be I I with respect to spot we're talking about 2500 spot robots out in the world doing inspection tasks doing public safety.

[00:10:45]>> So uh with the plans for Atlas you would see several times that level of deployment for Atlas. Um why why do you think u the best first use case for Atlas will be factory?

[00:11:01]>> That's where we think there's obvious value.

[00:11:05]>> Uh there's a shortage of workers that is on the horizon in the United States and of course here in Japan is already being felt. And so the the jobs that you want to automate are the ones that are dangerous or or potentially hazardous, lifting heavy things, uh moving things around in a factory. Those are great tasks for a robot >> and not so good for humans who who get hurt lifting things.

[00:11:32]>> Um so we think that the value is is there. Um, and that's what we're focusing on first because the robot is making something that then has value to another downstream consumer. Uh, there's lots of applications for humanoids. Um, particularly at at the summit, I'm hearing all kinds of amazing >> presentations um, including household applications, laundry, dishes, you know, things like that. I think those are very exciting.

[00:11:59]Um, the question will be what is the value to whom? And for us, having robots that help make other things is how you take a robot and generate more value from it downstream in an industrial context and also help address the workforce shortage. So, it just makes sense to us to go where the demand is in terms of the workforce shortage and also where the value creation is.

[00:12:24]>> Mhm. I think there are lots of kinds of robots out there in the world. Um, why do you think we need humanoids in the first place?

[00:12:34]>> And why why does it have to be shaped like human?

[00:12:39]>> It doesn't. And and we started with with Spot and then we moved on to Stretch, which doesn't have legs. That's a stretch is our warehouse box moving robot.

[00:12:48]>> Um, we did have Atlas as a research platform >> for the past uh like 13 years >> doing the back flips.

[00:12:57]>> Yeah. Yeah. doing the parkour, doing the flips. Um, and I think for a while we thought that's an interesting research platform, but is it really the form factor that we need for a robot? And I think over time as we saw spot deployed and being used mostly as an inspection or a hazard mitigation, we realized that spot can't quite do a lot of the tasks in places that are designed for humans because of you know reach because of the need to carry things with two hands because of the weight. the humanoid form factor made sense. But also in terms of data collection from humans in in this AI revolution for robotics, you need to generate operational behavioral data.

[00:13:44]And the best way to do that apart from simulation and other techniques is to have a human do the task with like a capture suit and to have that that data then used by the AI AI model and the robot to do the behavior.

[00:14:00]>> So I think that I think that was part of the turning point is the ability of AI to accelerate development >> within the human form factor. Mhm.

[00:14:10]>> Uh and I'm aware that uh Dustin Dynamics has a partnership with Google Deep Minds or Gemini model. Uh is it a important like a step for Atlas that the robots like Atlas to be implemented like in the real physical environment?

[00:14:28]>> It's a great partnership. We're doing a lot of exciting amazing work with uh with Karolina Prada and her team.

[00:14:34]>> Um I I think it's growing great.

[00:14:37]>> Okay. And you said the robots are living in factories uh earlier today, right?

[00:14:43]>> And how are you going to see the robots in like our everyday life or is that us even coming to the household?

[00:14:53]>> No. Uh so so we are focused on industrial applications as as the initial deployment for Atlas. Um as we discussed that's where we see the value for this kind of robot. But what's exciting and what I was talking about um in my presentation here at the summit is there are other companies that are focused on home applications that are amazing and and there's lots of potential there. uh it makes sense to to to pursue it all because again the human form factor is amendable to use everywhere and with the AI revolution you can train those robots >> using techniques from human operators >> so I I think it's an an exciting time uh it it just at the moment we are very focused on industrial applications >> okay um you also talked about dual use of robots and which means both privacy sector and public sector like militaries are can also use robots uh respectively and there's been a lot of discussion around uh military use of AI and so how are you thinking about what military use of robots should be well we think robots do a lot of good at keeping people safe from harm and uh we have for example a um a great customer the the customer of the Dutch Ministry of Defense and what they do is bomb disposal. They find unexloded ordinance, including from the Second World War, and that's the kind of thing you don't want to send a human to investigate. So, keeping people safe from harm, providing information about um you know, what what threat might be over the hill, uh are there hazardous substances in the environment? So using spot with the SEAB burn kit allows you to detect chemical, nuclear, biological, radiological, and nuclear substances that could harm people. And so our defense uh customers are really appreciative of the ability to keep people safe from harm uh and and to take action if there's something like a like an unexloded or or an IED, >> an imp, you know, improvised explosive device. We have a um a disruptor payload on spot. I I didn't show this on stage because it wasn't sort of topical to what we're talking about here. Humanoids mostly, but um that device is able to take a bomb and basically using uh using water just shred the bomb and disable it before it can explode.

[00:17:33]>> That's very useful to defense agencies.

[00:17:35]M >> do you think the humans humanoids like Atlas is going to be a real fighter in the military context or warfare?

[00:17:45]>> We we are very focused on manufacturing and industrial applications. I there there may be other companies that are chasing that type of thing but we are not. Mhm.

[00:17:55]>> But um I think the Boston Dynamics uh was being really developed from the uh projects like DARPA uh projects. So I think the the military use or the research uh did accelerate the research of robotics. Do you agree with that?

[00:18:17]>> Sure. uh DARPA funding helped develop the underlying technology for some of the legged robotics innovations over the years. Um although that all predates the AI revolution.

[00:18:29]>> Um but certainly that was important to the early history of of the company.

[00:18:34]Those applications were all non-weaponized. So they were focused on carrying gear for for troops or how do you get a robot to the top of a building? So we have, you know, our sandflea robot uh leaps to the top of the building. We had um robots that crawl up the side of a wall. Those are all uh funded experiments in robot mobility, >> not in weapon systems.

[00:18:58]>> Mhm. Okay. And since you're here in Japan, um what kind of commercial or collaboration opportunities are do you see here? uh the parent company of Bon Dynamics used to be software group which is a Japanese company I think that there's a lot of relationship between the company and Japan so how do you see that >> well I so I focus on public policy and government relations right that's my role with the company so what I'm most excited about is figuring out how we can bring the Japanese government and the US government together to work on robotic strategies which could be supply chain resilience um it could be uh defense applications like we just discussed. Um it could be collaborations uh on deployment. I think it's really important to figure out how to put the robots to use to address the labor shortage. So I I do think we have an important role here. We are still exploring um uh you know the the potential >> partnerships and and initiatives that we can do here.

[00:20:04]>> What do you think about the uh the Japanese policy around robotics?

[00:20:09]Uh I was very impressed. Um uh the um the representative from Mati who is here yesterday uh Kuya uh explained how in the new initiative from the Japanese government uh they are helping to develop foundation models and to generate the data that's needed by Japanese companies to develop robots.

[00:20:33]That's amazing. That's not typically something the American government would get so involved in. American government might fund research or fund um certain kinds of um uh research initiatives uh but not typically um in that way. Um so I thought that was very impressive and it as as he explained >> that came out of really the need for for that sort of foundation of AI robotics in Japan which doesn't otherwise exist.

[00:21:03]So I thought that was great to see and I I think all countries should have a national robotic strategy. In fact, that is the main message that I've carried in Washington over the past 3 years is to see the United States develop and launch a national robotic strategy, which is not something that that we currently have.

[00:21:20]>> Mhm. So, I think that the policy on robotics in the US uh is typically like funded research and sort of thing.

[00:21:30]Really the only policy for robotics in the US has been the national robotics initiative which was developed under President Obama which funded university research in robotics. That program has concluded. So really currently there's there is no uh federal robotics policy initiative or strategy or really any support in in the policy world for robotics and that's that's what we're trying to help change. M >> do you see any change uh after the second Trump administration because he uh announced AI action plan last year and did did it change uh also the policy on robotics >> in the AI action plan there's a section on advanced manufacturing that specifically refers to robotics and that calls on the NTIA which is the national telecommunications and infrastructure agency that um that's a part of the department of commerce and they convened the robotics industry to discuss robotics supply chain challenges um workforce education and training. Um so the answer to your question is is yes. We are seeing a lot of interest and engagement from the Trump administration on trying to learn about the robotics industry, figure out how they might help support us.

[00:22:51]>> Um, and also, you know, uh, putting robots to use and what are the ways in which government can help encourage or inspire industries to be more proactive when it comes to robot adoption.

[00:23:03]>> Okay, I have last question. Um >> what kind of future are you seeing uh specifically for the company Boston Dynamics or robotics industry as a whole?

[00:23:16]>> It's a really exciting and I think unprecedented time. I you know if you look at what's happened particularly in humanoids the past year or two tremendous funding rounds, tremendous uh amount of attention and publicity. I think it's it's super exciting. You know, we we've been doing this a long time and to some extent when we showed Atlas jumping and like people were like, "Okay, that's interesting, but what is that even about?" And now >> everyone is inspired to take these technologies, use AI, and create machines that really improve people's lives, help keep people safe, >> um address the workforce shortage here and in other countries. I think it's it's an amazing era to be in robotics and you know it's to us it's like a fulfillment of our vision over 30 years ago when our founder Dr. Raybert >> said, "Hey, robots are going to have a role in society doing all kinds of things. If we can make them walk, >> well, here we are with dozens of companies, maybe even hundreds >> with walking humanoid robots. I think it's a very exciting time."

[00:24:29]>> Okay, great. Thank you very much for your time.

[00:24:32]>> Thank you. Appreciate it. Yeah. Mhm.