Dynamic Obstacle Avoidance for Manipulators Based on Voxel Environmental Perception and Sequential Quadratic Programming

A dynamic obstacle avoidance algorithm based on Sequential Quadratic Programming (SQP) and voxel environmental perception is proposed to address the manipulator's avoidance of both static and dynamic obstacles in unstructured environments. During offline planning, the RRT-Connect algorithm is applied for static obstacles in joint space, with a pruning method to quickly generate the path. In the online phase, the obstacle avoidance problem for non-prior dynamic obstacles is formulated as an SQP optimization problem, solved in real-time at each time step using the offline path as reference. Simulation results show that the algorithm allows the manipulator to sense and avoid dynamic obstacles while following the offline path to reach the target pose.