Impedance Control and Reactive Self-Collision Handling for Humanoid Dual-Arm Manipulator Systems

Due to their high degrees of freedom, dual-arm robotic systems offer more versatility for executing complex, cooperative tasks. Yet they introduce new challenges in motion planning and control. In particular, they are able to touch and thus might collide with themselves. Therefore, the robot must be equipped with collision avoidance methods. This work presents an integrated, whole-body control framework for reactive, self-collision handling in a redundant dual-arm manipulator system. The architecture unifies a joint-space impedance controller with reactive collision handling algorithms. Specifically, this work investigates and implements an approach based on a state-dependent damping force. This force is derived from a reduced, one-dimensional model of the dynamics along the impact line connecting the two colliding points, and is dynamically modulated according to the system state (relative distance and velocity) as well as the effective reflected inertia of the approaching links. By dissipating energy, the proposed strategy slows down unintended human-guided motions and mitigates impact velocity. Thereby it enables controlled self-contact, a feature desirable in human-robot interaction such as kinesthetic teaching. Our method was verified in real-world experiments. The proposed framework relies on Pinocchio for the real-time computation of analytical kinematics and rigid-body dynamics, and was validated experimentally on the physical 15 degrees of freedom dual-arm hardware, with the base joint held fixed so that only the two arms’ 14 joints were actively driven throughout the validation. The experiments demonstrate that the proposed state-dependent damping strategy achieves enhanced robustness, an absence of undesired oscillations, and high trajectory tracking accuracy in the replay phase of kinesthetic teaching, without the strict avoidance constraint of classical repulsive potential methods.