Article: 基于轨迹调制和强化学习的小型四足机器人运动控制方法, Motion Control Method for Small Quadruped Robots Based on Trajectory Modulation and Reinforcement Learning
Patent: 黄凯, 张子韬, 赵子健, 陶若怡, "一种基于强化学习的小型仿生鼠机器人的运动控制方法"（申请号：2023116499786，2023）
Author: Zijian Zhao
Supervisors: Prof. Kai Huang, Prof. Gang Chen, Zitao Zhang
The official GitHub for the conference version is RS2002/ARS-Bezier: Official Repository for The Paper, ARS-Bezier: A Trajectory-based Reinforcement Learning Approach for Autonomous Locomotion of a Rat Robot (github.com).
To train the model, use the following command. You can also use --help to find useful parameters such as learning rate.
python train.py --modelPath <environment path>If you want to use the ARS-Bezier (simple) method, you can run train_simple.py in the same way.
To evaluate your model, you can use:
python eval.py --modelPath <environment path> --parameterPath <parameter path>or
python eval_simple.py --modelPath <environment path> --parameterPath <parameter path>To train the model in a random environment, you can use the random_env function in environment.py and set the --modelPath parameter as ./models/my_test.xml.
Other models like PPO and ETG-RL are located in the "discard" folder, but they may have some bugs.
