Wenhui Wang, Leqing Li, Fumeng Ye, Yumin Peng, and Yiming Ma
DDPG algorithm, reward function, deep reinforcement learning, underwater robot, large-scale path planning
To ensure the effect and improve the accuracy of large-scale path planning for underwater robots, a large-scale algorithm for planning the path for underwater robots based on deep reinforcement learning is proposed. Deep reinforcement learning is analysed, and the idea, structure, network update method, and training process of deep deterministic policy gradients (DDPG) algorithm are described. A fitness learning model of the robot which under water is confirmed to describe the mathematical relationship between the geographical location and operating speed of the underwater robots. On this basis, DDPG algorithm is applied in large-scale path planning of underwater robots. TensorFlow is used to build Actor and Critic neural network structures, and design environment state models, action state spaces, and reward functions. In deep reinforcement learning, the large-scale navigation planning for the underwater robot, through exploration-online trial and error, finds the optimal search strategy, and considers obtaining the maximum expected reward during the path planning procedure, achieving the large-scale path planning for the underwater robot. According to the experimental results, the proposed algorithm demonstrates good performance in large-scale path planning for underwater robots and effectively improves both the accuracy and efficiency of the planning process.
Important Links:
Go Back
