Online Static Obstacle Avoidance and Offline Static Obstacle Avoidance Framework Based on Interaction Probabilistic Movement Primitives

Abstract

How to make robots self-adaptive to obstacle avoidance in the process of human-robot collaboration is one of the challenges in the community. In an actual environment, robots often encounter unanticipated obstacles that make it difficult to complete a task. So we in this paper proposed an obstacle avoidance framework based on Interaction Probabilistic Movement Primitives (iProMP), which combines online static obstacle avoidance with offline static obstacle avoidance. For unanticipated obstacles in human-robot collaboration, we find obstacle avoidance trajectories by solving the Lagrange equation, and then the product of Gaussian distribution is used to fuse the two iProMP trajectories to smoothly switch from the original trajectory to the obstacle avoidance trajectory to achieve fast online static obstacle avoidance. However, the obstacle avoidance trajectory is not optimal. When human-robot collaboration is over, the obstacle is usually not immediately cleared, and the unanticipated obstacles become the anticipated obstacles. In order to obtain a better obstacle avoidance trajectory, Path Integral Policy Improvement with Covariance Matrix Adaptation algorithm is used to train the demonstration trajectories to obtain new iProMP parameters, using the new parameters of human-robot cooperation to realize offline static obstacle avoidance. Experimental results based on two-dimensional trajectory obstacle avoidance and UR5 obstacle avoidance demonstrate the feasibility and effectiveness of the proposed framework.