There are now robots that can adapt like animals

Play

Robots can now adapt like animals and humans when damaged.

YouTube / Nature

Robots that can learn and adapt are no longer features of science fiction. In the past, robots would function until they were broken. Then, someone would fix them, if possible. Now, robots are adapting to “injuries” so that they can complete their tasks.

Antoine Cully

Adaptation is an inherent trait to animals and humans but now, robots can be programmed to make changes in the way they operate if a part of them has been damaged. A team of robotics specialists at Pierre and Marie Curie University of Paris (UPMC) have developed such robots who have adapted or even “self-healed” during the course of their simulations.

On May 31, Gizmag pointed to a new paper published in Nature magazinewhich relates that robots would be used more outside of manufacturing such as in rescue and disaster response situations if it weren’t for their fragility.
Though robots are built to withstand harsher environments than humans or animals ever could, they can break down and when they do, they can be rendered useless. The only thing anyone has been able to do to prepare is to pre-program robots with responses to possible dangers. Doing so, however, is a bit impractical.

“A promising approach to reducing robot fragility involves having robots learn appropriate behaviours in response to damage,” writes the UPMC robotics specialists. In their trials, the specialists successfully tested robots that adapted without pre-set programs and they were able to do so within two minutes.

Before the robot is deployed, it uses a novel technique to create a detailed map of the space of high-performing behaviours. This map represents the robot’s prior knowledge about what behaviours it can perform and their value. When the robot is damaged, it uses this prior knowledge to guide a trial-and-error learning algorithm that conducts intelligent experiments to rapidly discover a behaviour that compensates for the damage.

Essentially, the robot learns or improvises in real time. In their simulations, the robotics specialists damaged a multi-legged robot in five different ways. Damages included broken or missing legs and broken joints in the robot’s arms. Specialists had implemented an algorithm that essentially programmed the robots to test via trial and error possible solutions to get their jobs done. The robots self-assessed in relation to their tasks, trying out solutions until tasks could be performed successfully. This video shows examples of how robots adapted to accomplish their tasks.

The new algorithm used with the robots tested in the UPMC trial can eventually lead to robots that are more self-sufficient. “Once tested, the robot becomes like a scientist,” lead author, Antoine Cully explains. The robot begins with pre-set “expectations” and when those expectations are not being met, the robot experiments with new behaviors until something works.

*originally published on the now defunct Examiner.com