Autonomous Decision-Making Generation of UAV based on Soft Actor-Critic Algorithm

Abstract

In this paper, we utilize deep reinforcement learning algorithm Soft Actor-Critic (SAC) to solve the autonomous decision-making problem of Unmanned Aerial Vehicle (UAV). Firstly, the decision-making problem of UAV is abstracted into a game scenario, and the simulation environment is built based on tensorflow, pygame, etc. Secondly, the decision-making problem of UAV is modeled as Markov Decision Process (MDP), the reinforcement learning framework is constructed, and the SAC algorithm is connected with the simulation environment. Finally, the model begins to be trained, so that the UAV learns the actions of the task, finds the optimal strategy, and generates autonomous decision-making capabilities based on SAC algorithm. It can take full advantage of reinforcement learning (RL) in sample-free learning. Facing the complex and unknown decision-making environment, the SAC algorithm can make the UAV continuously summarize experience in the process of interacting with the environment and make the best strategic choice. The effectiveness of autonomous decision-making based on RL is verified through simulation experiments.