Memory-based Stochastic Trajectory Optimization for Manipulator Obstacle Avoiding Motion Planning
Yunfan Wang, Xian Guo · 2022
Obstacle avoidance planning for manipulator is always a challenging research problem especially for manipulator with high degree-of-freedom. The Stochastic Trajectory Optimization for Motion Planning (STOMP) can find solutions efficiently by generating noise trajectories to optimize. However, STOMP’s optimization process starts with an initial trajectory linearly interpolating between the start and the goal configuration, which maybe far away from the optimal trajectory. In addition, STOMP cannot use past knowledge to accelerate the solution of new planning tasks, resulting in more iterations to reach a satisfactory solution. In this paper, We propose a Memory-based STOMP (M-STOMP) algorithm, which stores past trajectories in Memory so that it can generalize the experience obtained from past trajectory optimization to new and similar tasks. Before optimization, a trajectory most relevant to the current situation is matched and that trajectory must be modified in order to reduce the difference between the matched trajectory and the current task. In the optimization process, a more efficient way to generate noise trajectory is used to speed up the convergence process. After optimization, the highly divergent trajectories are stored in memory to maintain diversity and limit memory size. In the simulation experiment, we set up several new planning tasks with different obstacles and goal configurations to verify the effectiveness of the algorithm. The simulation results show that the proposed method can obtain collision-free trajectory with shorter iterations, and the algorithm has more advantages in the new similar planning tasks.