A multi-agent approach to hyper-redundant manipulators

The ability to reach into con??ded and unorganized envi- ronments, and grasp and manipulate various objects make hyper-redundant robotic manipulators the ideal tool for a variety of applications. Such applications include search and rescue, exploration of unknown environments, assem- bly and manufacturing as well as robotic-surgery. A key challenge that has limited the applicability of such robotic manipulators is the di??culty in controlling a robot with a very large number of interacting components. This paper aims to address this issue by using a new adaptive multia- gent control approach for exible shape changes of so called snake-arms and discusses extensions including a universal, more versatile, tree-like robotic manipulator. Our prelim- inary results show the feasibility of the approach and the potential bene??ts of a multiagent approach which includes scalability, fault tolerance, adaptivity and automated load balancing. A key ??nding of this study is the necessity of principled credit assignment for the agents, so that their collective actions optimize the performance of the full robot structure.