Wired UK: AMARSi project could see robots learn from co-workers
Another recent story for Wired:
Robots of the future will be capable of learning more complex behaviours than ever before if a new, pan-European research project succeeds in its goal of developing the world’s first architecture for advanced robotic motor skills.
If successful, the four-year AMARSi (Adaptive Modular Architecture for Rich Motor Skills) project (which started this month) could see a manufacturing world filled with autonomous, intelligent humanoid worker bots that can learn new skills by interacting with their co-workers. It could also see a society with personal carer bots capable of quickly adapting to complex environments and changing human needs.
If the researchers are successful, the 7 million euro, EU-funded project will enable humanoid (and quadruped) bots to autonomously learn and develop motor skills in open-ended environments in the same way humans do — by learning from the data provided by movement and essentially rewiring their circuits to process and store the new knowledge they’ve acquired.
It’s all a far cry from the limited learning and motor skills capabilities of existing bots and it will rely on a suitably advanced range of technologies to make it happen: dynamic neural networks built on reservoir computing principles, new robotics hardware designs, and sophisticated software algorithms are all involved.
AMARSi relies on a “more-or-less unusual,” biologically inspired view of motor skills that goes beyond traditional robotic designs and is better suited to truly autonomous robots, says Project Coordinator, Jochen Steil, Director of The Cognitive Robotics and Learning Laboratory (CoR-Lab), at Bielefield University, in Germany.
Read more about the AMARSi project here.