A method and system for safe visual servoing control of a robot arm based on reinforcement learning

Abstract

The application discloses a kind of mechanical arm safety vision servo control method and system based on reinforcement learning, belong to the intersection technical field of robot control and artificial intelligence. The visual feature information of target object in image plane is acquired using mechanical arm hand-eye camera;Define the state space of mechanical arm body state, visual feature state and obstacle observation, mechanical arm action space and position control mode;Sparse safety reward function is set, including vision servo success reward, visual feature point out camera field of view penalty and CBF inspired safety penalty;Sparse safety reward function is used to train mechanical arm to avoid obstacle servo movement, construct visual feature sequence dataset, and offline train gated recurrent unit GRU visual feature completion model, so that the target is missing for not more than 3 seconds in the process of mechanical arm obstacle avoidance servo movement, obstacle avoidance servo movement can still be realized to reach target object, the application can realize the stable control of mechanical arm vision servo in multi-obstacle and hand-eye camera short-time shielding environment, improve system robustness, and effectively reduce collision risk.

Images