A novel trajectory tracking algorithm based on soft actor-critic-ResNet model
Haohua Li, Jie Qi · 2023
In order to effectively accomplish the control of robot in unknown environments, we propose a soft actor-critic- ResNet(SAC-ResNet) model based on multi-sensor information. First of all, comprehensively consider issues such as obstacle avoidance and robot walking efficiency, we design a state space based on various sensors and a continuous action space. Secondly, SAC combined with ResNet, multilayer perceptron(MLP) and multi-head-attention realize the organic integration of multi-sensor information, effectively solving the problems of low efficiency and environmental dependence caused by insufficient environmental information. The simulation results show that the SAC-ResNet model based on multi-sensor information can continuously control the robot to avoid unknown obstacles effectively in different environments, improving the accuracy, safety and robustness of local obstacle avoidance.