Real-Time Self-Collision Avoidance in Joint Space for Humanoid Robots

In this letter, we propose a real-time self-collision avoidance approach for whole-body humanoid robot control. To achieve this, learn the feasible regions of control in humanoid's joint space as smooth boundary functions. Collision-free motions are generated online by treating learned functions constraints Quadratic Program based Inverse Kinematic solver. As geometrical complexity grows with number degrees-of-freedom (DoF), learning computationally efficient and accurate is challenging. We address partitioning model into multiple lower-dimensional submodels. compare performance several state-of-the-art machine techniques to such Our validated on 29-DoF iCub robot, demonstrating highly avoidance.