Toyota Research Institute SVP on the difficulty of building the perfect home robot • TechCrunch

early this week, Toyota Research Institute opened the doors of its Bay Area office to members of the media for the first time. It was his day with demonstrations ranging from driving his simulator and drifting his instructor to conversations about machine learning and sustainability.

Robotics, a longtime focus of Toyota’s research division, was also on display. SVP Max Bajracharya introduced two projects. The first was closer to what you would expect from a Toyota. An industrial arm with an improved gripper designed for the incredibly complex task of moving boxes from the back of a truck onto a nearby conveyor belt. This is what most factories want to automate. future.

The other is a little more surprising. At least for those who haven’t followed the department’s work that closely. A shopping robot picks different items from a shelf based on barcodes and general location. The system can extend to the top shelf to find items before determining the best way to grab various objects and put them in the basket.

The system is a direct outgrowth of the 50-person robotic team’s focus on elderly care and is intended to address Japan’s aging society. However, this represents a shift from his original work of building robots designed to perform household chores such as washing dishes and preparing meals.

You can read a longer coverage of that pivot in an article published on TechCrunch earlier this week. It is drawn from a conversation with Bajracharya and is printed in full below. Please note that the text has been redacted for clarity and length.

TechCrunch: I was hoping to get a demo of a home robot.

Max Bajracharya: We’re still doing some home robot stuff[…] What we have been doing has changed. Home was his one of our first challenge tasks.

Eldercare was the first pillar.

absolutely. One of the things he learned along the way was that we didn’t measure our progress very well. Home is very hard. Choosing a challenge task is difficult. The problem with the house wasn’t that it was too difficult. It was just too hard to measure the progress we were making. I messed up procedurally. We put flour and rice on the table and tried to wipe them off. I put things all over the house to keep the robot tidy. We were on Airbnb to see if it worked, but the problem was we couldn’t go back to the same house every time. It will overfit.

Wouldn’t it be ideal not to go to the same house every time?

You’re right, the problem was that we couldn’t measure how well we were doing. Let’s just say I got a little better at tidying up this house. I don’t know if it’s because we’re more capable, or because the house just got a little easier. We were like, “Show me a demo, show me a cool video. We’re not good enough. This is a cool video.” The grocery challenge task we mentioned requires an environment that is as difficult or has the same representative problems as home, but a place where you can measure how much progress you are making.

I’m not talking about specific home or supermarket goals, but solving problems that can straddle both of these locations.

Or even measure whether we’re pushing the cutting edge of robotics. Any challenge is fine. The DARPA Robotics Challenge, they were hard building tasks. The same goes for challenge tasks. We love our home because it represents where we ultimately want to help people at home. But it doesn’t have to be home. The grocery market is a great example because it is so diverse.

But there is frustration. I know how hard these challenges are and how far things are, but if a random person sees your video and you can’t tell it, suddenly it’s over the horizon It is in

absolutely.so gil [Pratt] every time I say, “Re-emphasize why this is a challenge task.”

How do you translate that to normal people? Ordinary people don’t get carried away with challenging tasks.

You’re right, the demonstration you saw today was intended to show a challenge task, but it also showed an example of how the functionality from that challenge can be applied to a real application, such as unloading a container. that’s a real problem. We went to the factory and they said, ‘Yes, this is a problem. From these challenges, we’ve shown some breakthroughs that we think are important, and we’re trying to apply them to real-world applications.

How big is your robotics team?

The department is about 50 people evenly split between here and Cambridge, Massachusetts.

There are examples like Tesla and Figures trying to make multi-purpose humanoid robots. You seem to be heading in a different direction.

Slightly. One thing we’ve observed is that the world is built for humans.If it’s just a blank slate, you’re saying I want to build a robot that works in human space. . You tend to end up with human proportions and human-level abilities. Not necessarily because it’s the best solution. Because the world is designed around people.

How do you measure milestones? What does success look like for your team?

Traveling from home to the grocery store is a great example. We were making progress at home, but it wasn’t as fast or clear as when we moved to the grocery store. It really reveals what the real problems are in your system.And you can really focus on fixing those problems. When I toured both Toyota’s distribution and manufacturing facilities, I saw all of these opportunities that were basically grocery shopping challenges, except that they were a little different. All parts are distribution center parts.

I’ve heard from 1,000 people that home robots are very difficult, but I feel the need to try it myself and make the same mistakes they did.

I think I’m probably as guilty as everyone else. It’s just that GPUs are better now. Ah, now that we have machine learning, we know we can do this. Oh, maybe it was harder than we thought.

Something has to flip it over at some point.

perhaps. I think it will take a long time. Just like self-driving cars, I don’t think there’s a silver bullet. It’s not some kind of magic like this, it’s “Okay, it’s solved.” It is scraped away little by little, and it is gradually scraped away. That’s why it’s important to have such a roadmap with a short timeline. Shorter milestones allow you to achieve smaller achievements. That way, you can keep working to actually achieve your long-term vision.

What is the actual process of bringing these technologies to market?

This is a very good question that we are trying to answer ourselves. I think we have some understanding of the landscape now. Maybe I was naive at first and thought, ok, I just need to find this person who is going to pass this technology on to a third party or someone inside Toyota. I think we’ve learned that whether it’s a business unit, a company, a startup, or a unit within Toyota, they don’t seem to exist. I’m trying to find a way, but I think that’s also what TRI-AD is talking about. It was created to translate the self-driving research we were doing into something more realistic. We have the same problem in robotics and many of the advanced technologies we work with.

I’m thinking about the possibility of getting to a place where we can spin off.

potentially. But it is not the primary mechanism for commercializing the technology.

What are the main mechanisms?

don’t understand. The answer is that the diversity of what we do is very likely different from group to group.

How has TRI changed since its founding?

When I first started, it felt like we were obviously doing robotics research. One reason is that the human environment is far from a technology that can be applied to almost all difficult real-world applications. Over the last five years, we feel like we’ve made enough progress on this very difficult problem that we’re starting to see it translate into real-world applications. We made a conscious shift. We are still 80% ahead of the curve in our research, but to determine if the research is as good as we think it is, and whether it can be applied in practice, our resources allocates perhaps 20% of – World application. You may fail. You may find that you thought you had some interesting breakthroughs, but they aren’t reliable or fast enough. But we spend 20% of our effort trying.

How does aged care fit into this?

In a way, it is still our North Star. The project is still looking at how we can ultimately amplify people at home. When it comes out bit by bit, we use these short-term milestones to show progress in the research we’re doing.

How realistic is the possibility of a completely extinguished element?

In the future, if it becomes possible to start from scratch, I think there is a possibility. If you look at manufacturing today, especially Toyota’s manufacturing, it’s unlikely to come close.we [told factory workers], we are building robotic technology, where do you think it could be applied? They showed us so many processes. For example, run this wire harness here, pull it out here, clip it here, clip it here, take it here, get it here, do it like this. And this takes five days to master the skill.

But the hardest thing for people is what they want to automate.

Yes, it can be difficult or injury prone. Sure, we’d like to eventually build a foothold to get there, but with today’s robotics technology, we’re nowhere near that.