Chris Atkeson: The Roboticist Who Inspired Disney
How did your work influence Disney’s Big Hero 6 and the character of Baymax?
Around that time, Disney had just bought Marvel, so Don Hall, a director with Walt Disney Animation Studios, started to look into, “Okay, what does Marvel have that we could do something with?” For some reason, they got very excited about the original Baymax, a giant metal monster. And when Disney makes a movie, they do research. For Frozen, they sent a bunch of animators from Southern California to Norway to play in the snow.
For Baymax, they said, “We don’t really want a terminator robot or something that’s scary. What are our options?” They arranged a tour to a bunch of robotic robot labs across the world and one of the places they visited was [Carnegie Mellon]. We said, “We think robots in the future will be very light. Our version of that is inflatable. A big application is taking care of people health wise.”
They grabbed onto those two ideas. But they did not reveal in the slightest way, no twitch of a facial muscle, no enlargement of the pupils that they were interested.
It was like talking to people with the CIA. They don’t say yes, they don’t say no, they just listen. By the end of the visit, they said, “Okay, could you sign a little something, so we don’t have any legal disputes after this.” They then hand me something that’s like two inches thick and said, “Sign here.”
But when they were done and the movie came out, they were incredibly generous in giving us some credit, which is extremely rare in the movie business. I was happy.
I read somewhere that the shape was inspired by a Japanese rice cooker, is that true?
There were all sorts of claims. My students believe that the shape of Baymax derived from looking at me. Who knows?
Where do you get your inspiration?
Inspiration comes from everything around us, they come from your life.
The Baymax idea was driven by this question of how to make it safe. From a physics point of view, if you’re light, you can’t store a lot of energy, you can’t have a lot of momentum, and you don’t generate a big force from gravity. It’s an incredibly simple idea.
I also spend a lot of time in toy stores.
How did you get interested in robotics?
There are two answers to that question: how I got into it and why I was receptive to it.
I come from a family of eight children, seven of whom were boys and one sister. They were all much more athletic than I was and some of them, later in college for example, set school records for various things. I, on the other hand, was very clumsy.
Moving, activity and being skillful was a big deal for me and in some sense, that’s what robotics is.
In terms of robotics, what do you focus on?
The central mystery is this: How do we get anywhere close to human levels of performance in physically interacting with the world?
It’s not easy and at this stage of the game, we, as very smart human beings, have to carefully craft both program, software and hardware to get anything to work. It’s very brittle. If something changes, everything fails.
For a long time, I’ve been interested in humanoid robots; robots that are trying to do what people do and getting them to do dynamic tasks. Right now, I’m working on robot skin and how to make robots have the reasoning capabilities that we all have.
What’s robot skin?
I’m mostly interested in the sensing part.
How do you know what you’re touching? There’s an experiment you can do. What you can do is you can anesthetize somebody’s hand so they can still move it, but they can’t feel anything. It’s almost impossible to do any kind of manipulation under those conditions.
What kind of materials or ideas are you experimenting with?
The mechanical properties of skin are very important. For a long time, robotics was interested in, essentially, fingers that are metal. You can’t do squat with fingers that are metal. There’s no friction.
When you touch something, you get a circle touch. There’s a finite area, the ridges towards the edges of that area start to slip first, and if you detect that, you’ll press harder. You have about a million little sensors in your skin. When we build a robot sensor, it’s kind of clumsy.
How can we possibly build something that has a million sensors? What do you know of that has a million sensors?
Eyeballs?
How can we turn feeling into vision? It always struck me as stupid that we only have two eyeballs. Why not put eyeballs on your fingers? Or on your palms? Or all over your body?
The Greeks had a mythical figure named Argus who had 100 eyeballs all over his body. The Japanese also have a spirit that guards temples with 100 eyeballs.
How would you protect those eyeball sensors?
We put them underneath the skin. But they’re eyeballs, so you can’t block the vision. Either you got to have windows, or you’ve got to have glass, or something else that’s transparent.
What if the eyeball sensors get dirty?
We can make transparent silicone, but that’s a magnet for dust. Instead, we have behaviors to clean things, just like wiping glasses.
One of the things I’m looking at now is chopsticks. How do you align a pair of chopsticks? We’d have to engage our vision system, think about it and move our fingers around. Instead, tapping against a surface to align them doesn’t require sense or thinking.
For a long time in high jumping (the sport), people thought the best way over was stomach down. That’s what the coaches told you to do. Then there was this guy named Fosbury who decided the best way to go over the bar was to have your back down.
He didn’t break any world records, but the people watching him said, “Wow. That’s a better way to do the thing.” Through the magic of learning, they all adopted that strategy and right away, the world record jumped up.
Are you specifically programming the details of behavior?
We try to write programs that can learn lots of different behaviors. Most of the things you do, somebody showed you how to do it first. That’s called cultural knowledge. So far, we’ve been expecting robots to figure things out by themselves. That’s called the ‘Russian water baby treatment’. You take a baby, you throw it in the water and say “swim”.
That’s not going to work. You got to show the baby how to swim, how to eat with a fork, how to wash dishes or clean the house.
What are the limitations to that?
Right now, robots are dumber than stones.
How’s the robot going to figure out of all the gazillion things going on in the world what mattered to complete a task. Did today’s temperature matter? Did today’s humidity matter? Most people will say no but turns out humidity matters a lot more than you think. Your skin changes its mechanics based on the humidity. Sometimes it gets so bad it dries out, your skin cracks.
Robots don’t have that biological component.
But the world does. Have you ever taken your laundry out of the dryer and it’s easy to pull apart, easy to fold versus full of electrostatic? That’s humidity. It completely changes how you have to do the thing.
What’s the real-life application for something like that?
You’ve been using robots for years: vacuum cleaner, dishwasher, washing machine and dryers. In a domestic life, that is a big deal.
A lot of time and energy goes into that. Now, the question is, do you want to have smarter robots? More capable robots? Many people say yes and the application of a robot that knows what’s going on is they’re able to do exactly what you tell them to do.
Would people be receptive to having a robot living with them?
People already have relationships with their machines. You have a car?
Yes.
Does it have a name?
Big Bertha.
Case closed. You have a relationship with a piece of metal, and if that somehow became more human like, you’d probably have a deeper relationship with it.
How close are we to having robotic helpers in our homes?
Right now, robots are incredibly slow.
The reason they’re slow is they have to look at everything. They have to very gingerly touch it, then they do something, then very carefully un-touch it.
If you go to a candy factory at some point, it is amazing. They just shoot this chocolate out. They shoot this roll of paper out. It’s all going by at 10 chocolate bars a second. Then it all gets folded in chocolate wrapped up. Boom. It’s super-fast. Why? Because they found a good way to do it where they didn’t have to sense, and it was very robust.
Do you think fictionalizing robots in films and TV has adjusted people’s expectations on what they can do?
If a robot is able to do anything at all, people assume it can do everything. But that’s not the movie’s fault. That’s because the way humans are wired.
A very basic determination that you as a human make really fast is, is that thing alive? Is it animate or not? Because if it’s animate, you potentially have to defend yourself or you want to eat it. If it’s not animate, you can take all the time in the world.
When we see something animate, we don’t say, “Whoa, that’s 10% animate.” We say, “That’s 100% animate.” So when we see something that looks like it’s alive, we assume it can bite us. It has evil thoughts.
It has the full repertoire of a human and it’s only with a great deal of familiarity that you can say, “Oh yeah, I love snakes, they don’t bite me.”
Can you mitigate that response by tweaking the robot to make it look more human?
Uncanny valley. If you almost get it right, but not quite, people hate it.
What’s your gutsiest high-risk-of-failure project?
I would say it’s one of the ones I’m doing now, trying to make robots smart. Why is that gutsy? Because lots of people over the last 50 years have tried to do the same thing and robots are not smart yet.
Has big tech involvement changed the robotic industry?
Around 10 years ago, there was a robotics industry, but it was very simple: robots installed in factories to do the same thing over and over again. There was essentially a great deal of isolation between the academics and the companies.
But in the last 10 years, particularly because of self-driving cars, that’s gone away. Now it’s pretty easy for roboticists to get jobs in the industry with salaries much, much higher than they would get in academics. They can join industrial research labs and have the similar freedom to what they’d have in a university.
Of course, that also adds to the number of people who want to do robotics. I actually think we’re going to revolutionize society!
All kinds of things are happening. Many of the big projects are happening in DeepMind, Google X and Facebook AI. There’s also OpenAI. You can do a lot more there than you can at a university.
What’s your biggest takeaway when trying to build a robot? Does it teach you something about the human condition?
Most people don’t notice how hard everyday life actually is until they try to build a robot that does it.
What’s the principal motto you live by? What’s something that’s a through line in everything you’ve ever done in your 61 years of living?
I think it’s really important to think about things in ways that other people haven’t thought about them before. To look at them in a different way. There’s a cliche for that: thinking outside the box.
What do you like to do in your spare time?
To be left alone, to just play with all my toys. That’s what makes me happy.
Do you have kids?
No. I have robots and students. I have fish.
Do you learn anything from the fish?
They know when I’m coming out to feed them.
What are you most excited to see in the world, in the next five, 10, 15 years, 20 years?
For years, I said I was going to make a robot to help my mother as she got older. Now I’m worried about myself as I get older. Now I say I’m going to build a robot to help take care of me, and I have a limited amount of time to do it. I’m very excited about that.
