If you think building a robot is hard, imagine teaching one. All of that effort that goes into programming it is just horrendous.
You see, to build a robot to open a door knob, per se, there’s essentially only 2 ways to teach it. The first is by motion planning, in which the programmer codes and specifies each of the motor’s movements, and the second is Mimicry in which robots learn by watching a human perform the task.
Both methods have disadvantages. Motion planning can allow you to code for all the different situations and variations the robot can encounter, but it takes a lot of time and effort. Mimicry is fast and simple, but blocks its adaptability, as one slight change of detail means that the robot has to learn everything from a scratch again.
This is why a team of researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have invented a method that combines the two methods together for the best results. Essentially, it’s a robot that teaches a robot “how to robot”. Nice.
Source: Youtube, MITCSAIL
The technology is called C-LEARN, and it works by making a robot learn a task through mimicry, then translate that data into code that other motion planning robots can use. Basically, C-LEARN is a robot that learns a task from a human, and then codes that task into an algorithm to teach it to other robots.
“By combining the intuitiveness of learning from demonstration with the precision of motion-planning algorithms, this approach can help robots do new types of tasks that they haven’t been able to learn before,” says researcher Claudia Pérez-D’Arpino.
Source: MIT News
This makes motion planning much easier. Instead of manually programming every single possible variety of movement and change in detail, the programmers can instead rely on a single shared library. The movements that C-LEARN also makes aren’t hard-coded, so the robots it teaches will be more adaptive to changes in detail. Their next step is to improve the algorithm even further by making it handle even more drastic changes in detail and circumstance.
It’s a pretty neat piece of tech.