Research on robot path planning based on deep attention Q-networks
Aiming at the problems that traditional robot path planning algorithms have slow convergence speed and low accuracy in partially observable environments,a robot path planning method based on deep attention Q network(DAQN)is proposed. Firstly,in order to overcome the limitations of the traditional DQN due to the lack of memory units when processing partially observable Markov decision processes (POMDP ),an improved DQN algorithm that fuses the attention mechanism is proposed to fully utilize and mine perceptual information including historical data. Secondly,a reward mechanism for movement distance and direction is of the robot designed based on the artificial potential field(APF)method to improve the accuracy of path planning. Finally,the effectiveness of the DAQN algorithm is verified in a two-dimensional grid map simulation environment. The results show that the path planning performance of the DAQN algorithm in partially observable environments is significantly better than other algorithms,the algorithm can achieve superior path planning effects in complex environments.
万方数据
维普
