Aiming at the redundant manipulator operation task that needs to ensure the end-effector trajectory tracking as much as possible in the dynamic obstacle scene, a loose null-space obstacle avoidance (LNOA) method based on reinforcement learning (RL) is proposed. Firstly, the joint motion is decomposed into trajectory tracking motion and loose null-space obstacle avoidance motion, and the latter is further decomposed into joint null-space motion and end-effector slack motion; on this basis, LNOA framework for obstacle avoidance is designed. Secondly, the RL method is introduced to learn the loose null-space obstacle avoidance motion generation strategy, so as to generate the end-effector slack component and joint null-space component autonomously, which is then combined with the trajectory tracking component to realize obstacle avoidance and end-effector trajectory maintenance simultaneously. Finally, the simulation is conducted to verify the effectiveness of the proposed LNOA method.