A Path Planning for Unmanned Aerial Vehicles Using SAC (Soft Actor Critic) Algorithm

Abstract

Path planning is an essential element in the autonomous flight control of unmanned aerial vehicles, where it is important to quickly establish the path in uncertain environments and avoid collisions with the terrain and obstacles. In particular, research and development of fully autonomous flight is necessary in the case of unmanned aerial vehicles performing search, reconnaissance, and detection in terrain where human intervention is difficult. This paper proposes a path planning design method using machine learning. It has the advantages of fast calculation speed and high repeatability in a two-dimensional environment. Using the Soft Actor–Critic (SAC), an algorithm based on reinforcement learning, research into machine learning, observation status, behavior, and reward functions are required to generate global paths. Additionally, the learning and path generation results are analyzed by conducting a learning-based path planning simulation in an environment with dynamic obstacles.