Dynamic Obstacle Avoidance of Fixed-wing Aircraft in Final Phase via Reinforcement Learning

Version 2023.11

Requirement

X-plane 11 simulator Pytorch 1.10.0 python 3.8 numpy 1.18.5 gym 0.10.5 pandas 1.3.5

Introduction

The codes provide a Reinforcement Learning (RL)-based obstacle avoidance strategy for aircraft. The RL-based obstacle avoidance strategy is based on the SAC RL algorithm and includes a pre-training stage and a fine-tuning stage. The codes also include a baseline 3DVO-based obstacle avoidance strategy. The codes require X-plane 11 flight simulator, Pytorch (1.10.0), Python (3.8), Numpy (1.18.5), Gym (0.10.5), and Pandas (1.3.5).

Files

SAC RL algorithm

• SAC.py - Establish an RL-based navigator according to the SAC RL algorithm
• ReplayBuffer.py - Establish a replay buffer

pre-training stage

• main_pretrain.py - The main file of pre-training
• main_pretrain_test.py - Test a pre-trained RL-based navigator
• Env_pre.py - Environment for pre-training and testing a pre-trained RL-based navigator

fine-tuning stage

• main.py - The main file of fine-tuning
• env.py - Environment for fine-tuning
• main_test - Test a fine-turned RL-based navigator
• Env_test.py - Environment for testing a fine-turned RL-based navigator

3DVO-based obstacle avoidance strategy

• main_test_vo - Test the 3DVO-based obstacle avoidance strategy
• Env_test_vo.py - Environment for testing the 3DVO-based obstacle avoidance strategy