介绍每一种强化学习算法的
- 算法原理 ✔️
- Pytorch实现 ✔️
| 单智能体强化学习算法 | Policy | Based |
|---|---|---|
| 👉 Sarsa | on-policy | value-based |
| 👉 Q-Learning | off-policy | value-based |
| 👉 DQN | off-policy | value-based |
| ❌ Rainbow-DQN | off-policy | value-based |
| 👉 REINFORCE | on-policy | policy-based |
| 👉 actor-critic | on-policy | policy-based |
| 👉 A2C | on-policy | Actor-Critic |
| 👉 DDPG | off-policy | Actor-Critic |
| 👉 HER-DDPG | off-policy | Actor-Critic |
| 👉 TD3 | off-policy | Actor-Critic |
| ❌ TRPO | on-policy | Actor-Critic |
| 👉 PPO-Continuous | on-policy | Actor-Critic |
| 👉 SAC | off-policy | Actor-Critic |
| 👉 Relay HER (RHER) | off-policy | Actor-Critic |
| 👉 Behavior Cloning (BC) | off-policy | Imitation Learning |
| 👉 Generative Adversarial Imitation Learning (GAIL) | on-policy | Imitation Learning |
运行环境：
python(in Pycharm)- 3.10
gymnasium-0.28.1
numpy-1.24.3
torch-2.1.0
建议使用 Pycharm，vscode或终端启动会有路径问题。建议更新 gymnasium 和 pytorch 到最新版本。算法原理请参考每个算法文件夹内的 markdown 文件，内部实现参考以算法名称命名的 .py 脚本，运行 train.py 脚本以进行训练。
在部分算法中添加了 tensorboard 模块，启动训练后会在对应算法文件夹内生成 log 文件夹，通过下面的终端命令可以打开网页查看训练日志：
tensorboard --logdir .