Watch the little robots get to know Swole in this virtual ‘gym’
If I ask To design a robot that is perfect for throwing blocks, you can think of something human, with legs for stability and hands for grip. And who can blame you? If humans are good at anything, it’s throwing things.
There’s a 0% chance you’re thinking of the thing in the video above, a Frankenstein’s monster that looks like pieces of Tetris put together. That’s because a computer “evolved” the robot’s body and brain in a new platform from MIT’s Computer Science and Artificial Intelligence Laboratory. It is called Evolution Gym, where instead of relying on human designers — and their human biases — the robots of tomorrow can hand their designs over to algorithms. “There’s the potential to find unexpected, new robot designs, and it also has the potential to create more high-performance robots,” said MIT computer scientist Wojciech Matusik. “If you start from very, very basic structures, how much intelligence can you actually create?”
A lot, as it turns out. Above you will see a soft robot learned how to do flips. Each “voxel” or color-coded unit in this robot serves a purpose. Black means a rigid part and gray means a software, neither of which has a positive effect. In contrast, voxels whose color is actuatoror bits of any robot that produces motion. (In a typical rigid robot, it’s the motors that power its joints.) Blue means the actuator contracts or expands vertically, while orange means it works horizontally. So when you watch that robot flip, it’s essentially spinning on the hard black voxel at its center, while the colored actuators push it off the ground.
In contrast, this is a simple domed robot made of only horizontally driven voxels. Basically the legs. But by contracting and expanding in coordination, the voxels make the machine move with surprising grace — a kind of gallop indeed.
This is one of learning how to climb mountains. Note the blue actuators on either side of the base, which alternate their movements to purchase items across the surface while a sort of soft robotic head appendage feels like walking up the pole. Another challenge is that parts of the column are soft, so the machine has to adapt to these when it wobbles in the upward direction. This is very complex behavior for such a simple robot, much less than a self-designed robot.