A 2D UAV Path Planning Method Based on Reinforcement Learning in the Presence of Dense Obstacles and Kinematic Constraints

The complex kinematic constraints and dense obstacles are always the huge challenge in the UAV path planning. To effectively deal with dense obstacles and kinematic constraints, a novel two-level optimization algorithm for unmanned aerial vehicles (UAVs) in 2D maps, called as Spherical Expansion-Proximal Policy Optimization (SE-PPO), is proposed in this paper. This method is a combination of SE and PPO algorithms. In the first level, SE algorithm is used to generate the initial path, and sub-goals are selected from this path in the first level. These sub-goals are optimized by the local path optimizer based on PPO algorithm to obtain the final path. The effectiveness of this method to deal with the kinematic constraints and dense obstacles is demonstrated by the results of the simulation experiments.