The new exciting project has been launched, with 7 million euro funding from EU, as a part of our pan-european cognition and robotics strategy. AMARSi or Adaptive Modular Architecture for Rich Motor Skills, is a 4-year long interdisciplinary research, which aims to develop a new breed of robots, that will learn motor skills by constant interacting with human counterpart. The idea is to combine dynamic neural networks build on reservoir computing, and new approach to hardware design. Robots will be interacting with human worker using strategies of imitation, but the human caretaker will be able to physically correct robot actions at any time, giving him more direct feedback for learning, without time-consuming reprogramming. It's like when you are showing your kid how to do things, those robots will be intended to work in a very similar way. The design will be based on two existing platforms: infant robot iCub (RobotCub) and four-legged Cheetah (Biorobotics). So to summarize - AMARSi is more about developing a software for a new learning strategies for robots, rather than redesigning hardware. Nevertheless, creating the motor skills architecture to enable such open-ended behaviour is a real challenge, so I am looking forward to see how they manage to achieve that.[photo credit: RobotCub]
No comments:
Post a Comment