Aiming at the problem of autonomous obstacle avoidance of dam inspection robot, an obstacle avoidance method based on Improved Deep Deterministic Policy Gradient (I-DDPG) algorithm is proposed to realize the inspection robot path planning in complex dynamic environment. In order to improve the convergence speed and stability of the algorithm, the prioritized experience replay mechanism is introduced to process the training samples, and also the random noise is added to improve the exploration ability of the algorithm. For the lack of a priori knowledge of the environment in the DDPG algorithm, the gravitational field function method is used to improve the fast convergence and the learning efficiency of the initial stage of the algorithm. The state action space, reward function and obstacle avoidance training process of the improved algorithm are designed. The simulation results show that there are much more advantages while the improved algorithm is employed, they are fast convergence speed, short path length and better obstacle avoidance performance for inspection robot who can successfully avoid dynamic obstacles under unknown complex environment. Through practical application, autonomous dynamic obstacle avoidance in complex environment is realized.