Adaptative Reaching Model for Visual-Motor Mapping Applied to Redundant Robotic Arms

Most of control algorithms for robotic reaching and grasping tasks, from visual and motor perception systems, are based on feedbacked systems. It supposes a limitation for the performance of remote reaching applications and for the robustness of the system. In this paper, a very robust learning-based model for visual-motor coordination is presented. This architecture is based on how the human system projects the sensorial stimulus onto motor joints and how it sends motor commands to each arm in open-loop mode, starting from initial visual and propioceptive information. The self-organizing characteristics of this model have allowed to obtain good results on robustness, flexibility and adaptability in both simulation and real robotic platforms. Coordination of the information from different spatial representations is based on VAM (Vector Associative Maps) algorithms, developed in CNS (Boston University). Indeed, compatibility requirements and system adaptation capability, give a solution for control of redundant systems.