To realize the autonomous obstacle avoidance and navigation of live working manipulator

we proposed a kind of subtask hierarchical deep reinforcement learning（HDRL） based on manipulator intelligent control model

and designed the corresponding environment state space

action strategy

and reward function. Meanwhile

we performed simulation experiments to verify the effect. The simulation results show that the success rate of obstacle avoidance and navigation model trained by single DRL algorithm is less than 35%. While the HDRL model not only gets better training convergence

but also has significant improvements in success rate which are 90% in wire stage and 71.01% in device stage. Moreover

the proposed safe operation rewards function can effectively guide an optimal route for manipulator considering the safety distance. The HDRL model keeps generalization performance in stage with different targets

providing theoretical and technical references for autonomous operation of live working manipulator.