Teaching Robots to Teach Robots

There are two ways to teach a robot to do something: Demonstrate the task and make it replicate it, or program the task directly into its software. The first method usually means the robot is limited to performing a very specific task and isn't able to transfer its skill to different situations, while the second method is usually time-consuming and requires skilled programmers to carry it out.

But what if it was possible to combine these two techniques? Researchers at the Massachusetts Institute of Technology (MIT) have developed a system that bridges both methods, enabling robots to transfer skills and knowledge amongst themselves. While it may sound like the beginning of a robot takeover, the new approach may prove useful in urgent situations, such as bomb disposal and disaster response.

The C-LEARN technique—described by the researchers in a paper recently accepted to the IEEE International conference on Robotics and Automation (ICRA)—involves providing a robot with a knowledge base of information of how to perform a task. A 3D interface is then used to show the robot a single demonstration of a specific task, allowing it to understand the motions it is being taught in the real world.

"This approach is actually very similar to how humans learn in terms of seeing how something's done and connecting it to what we already know about the world," said Claudia Pérez-D'Arpino, a PhD student who has written about the C-LEARN technique. "We can't magically learn from a single demonstration, so we take new information and match it to previous knowledge about our environment."

"Having a knowledge base is fairly common, but what's not common is integrating it with learning from demonstration," said Dmitry Berenson, an assistant professor of computer science at the University of Michigan, who was not involved in the research.

"That's very helpful, because if you are dealing with the same objects over and over again, you don't want to then have to start from scratch to teach the robot every new task."

A robot is therefore able to figure out how to perform a task even if there are obstacles in the way, as it does not simply learn one specific way to perform the action.

"It's good for the field that we're moving away from directly imitating motion, towards actually trying to infer the principles behind the motion," Berenson said. "By using these learned constraints in a motion planner, we can make systems that are far more flexible than those which just try to mimic what's being demonstrated."