A team of researchers at the Sorbonne University in Paris has created an algorithm that when added to the code of walking robots, teaches an injured robot how to walk again in just a few seconds. In their paper uploaded to the Preprint server arXiv, researchers- Antonie Cully, Jean Baptiste Mouret and Jeff Clune explain how their new technique works and how it will be helpful to the robots in future.
Unlike humans and animals, robots don’t perform well when something goes wrong in them. Dogs and cats for instance are suffering from na injured limb, they learn to hobble remarkably quickly. A robot with a damaged leg on the other hand is generally immobilized. The team decided to train an injured hexapod to experiment different walking styles through “Intelligent Trial and Error.” The earlier version of this algorithm kept the robots immobilized for about 20 minutes until they figured out how to move in a relatively straight line after suffering damage. The new technique lets the robot recover in about two minutes.
As you watch the video, the robot tries different waking configurations. First it hops a little, then it raises its damaged leg high to balance itself. At the same time it is alerted about its position in the room as well as its speed. If the style it has chosen is suitably straight and fast, it can keep trying until it gets moving again.
Well in this case the researchers created a database of potential styles-13,000 in all- by running a simulated version of the robot and rendering the walking styles. They also experimented with different types of injuries as well, from small to completely disabled legs, no matter which leg was disabled.
The new technique would be helpful to robots used in warfare.
Via : bbc