The skeleton algorithm for self-collision avoidance of a humanoid manipulator

For use in unstructured domains, highly redundant multi-arm robotic systems need both deliberative and reactive control schemes, in order to safely interact with the environment. The problem of collisions is crucial. A robust reactive algorithm, named the "skeleton algorithm", is proposed for the real-time generation of self-collision avoidance motions, where only proprioceptive sensory data are needed. The algorithm is applied to the DLR humanoid manipulator Justin, and a joint-torque control is used, where the collision avoidance torques are summed to the desired torques corresponding to other tasks; experimental results are reported.