And we asked a surgeon to perform a task — with the robot. So what we’re doing is asking the surgeon to perform the task, and we record the motions of the robot. So here’s an example. I’ll use tracing out a figure eight as an example. So here’s what it looks like when the robot — this is what the robot’s path looks like, those three examples.
Now, those are much better than what a novice like me could do, but they’re still jerky and imprecise. So we record all these examples, the data, and then go through a sequence of steps. First, we use a technique called dynamic time warping from speech recognition. And this allows us to temporally align all of the examples. And then we apply Kalman filtering, a technique from control theory, that allows us to statistically analyze all the noise and extract the desired trajectory that underlies them.
Now we take those human demonstrations — they’re all noisy and imperfect — and we extract from them an inferred task trajectory and control sequence for the robot. We then execute that on the robot, we observe what happens, then we adjust the controls, using a sequence of techniques called iterative learning. Then what we do is we increase the velocity a little bit. We observe the results, adjust the controls again, and observe what happens. And we go through this several rounds.
And here’s the result. That’s the inferred task trajectory, and here’s the robot moving at the speed of the human. Here’s four times the speed of the human. Here’s seven times. And here’s the robot operating at 10 times the speed of the human.
So we’re able to get a robot to perform a delicate task like a surgical subtask, at 10 times the speed of a human. So this project also, because of its involved practicing and learning, doing something over and over again, this project also has a lesson, which is: if you want to do something well, there’s no substitute for practice, practice, practice. So these are four of the lessons that I’ve learned from robots over the years. And the field of robotics has gotten much better over time.
Nowadays, high school students can build robots, like the industrial robot my dad and I tried to build. But, it’s very — now. And now, I have a daughter, named Odessa. She’s eight years old. And she likes robots, too. Maybe it runs in the family. I wish she could meet my dad. And now I get to teach her how things work, and we get to build projects together. And I wonder what kind of lessons she’ll learn from them.
Robots are the most human of our machines. They can’t solve all of the world’s problems, but I think they have something important to teach us. I invite all of you to think about the innovations that you’re interested in, the machines that you wish for. And think about what they might be telling you. Because I have a hunch that many of our technological innovations, the devices we dream about, can inspire us to be better humans.