Combining inverse kinematics and rigid constrained dynamics for online collision avoidance

Robots must compute collision-free trajectories on the fly to move proficiently in complex environments, including managing changes over time. In addition, the planner should be able to precisely reproduce the intended trajectory. State-of-the-art local planners are often more conservative in prioritising collision avoidance, typically sacrificing motion accuracy. Here, we propose a novel approach that combines closed-loop inverse kinematics, rigid constrained dynamics, and optimal control to overcome such a limitation. We evaluated its accuracy in simulation, comparing the motion generated near obstacles with that of a state-of-the-art algorithm. while we evaluated its robustness on a real-world bimanual teleoperation platform, including two anthropomorphic grippers. The results suggest that our algorithm can offer a competing alternative in collision-safe sampled motion tracking by achieving superior motion accuracy. In teleoperation, our method proved its efficiency in challenging real-world scenarios.