MIT teaches robots to train other robots

Robots have reached the next step in their evolution, thanks to researchers at the Massachusetts Institute of Technology. The Computer Science and Artificial Intelligence Lab has trained robots to teach each other new skills. Right now, these lessons are limited to relatively simple tasks such as opening doors and moving objects. MIT gives the example of a robot instructing another how to position its arms to attach a tire to a car. Once a robot learns a new skill, it can in turn do the same to other robots.

To pick up a new skill, the trainee robot does not need to be the same model as the trainer. This could make the C-Learn program applicable in a wide variety of settings. Researchers hope this could be used to quickly pass on skills to large groups of robots after a person teaches a skill to a single robot.

The potential impact of this technology could be huge if it matures into a more sophisticated program. For now, robot-to-robot training is time-consuming. It could take 30 minutes for a robot to teach another how to pick up a box. There are also all kinds of sophisticated tasks that cannot be taught yet.

Eventually, though, it is easy to imagine how this technology might be used to help commercial robots improve over time, possibly enabling them to become more efficient after just one robot learns a new skill. This could also be useful for service robots. Consumers, however, should be careful about which robots they let into their homes. They would not want their home assistant to pick up any bad habits from that delinquent robot from across the street.