当前位置：首页 > news >正文

具身系列——Double DQN算法实现CartPole游戏（强化学习）

news 来源：原创 2025/5/5 9:40:25

完整代码参考： rl/ddqn_cartpole.py · 陈先生/ailib - Gitee.com

部分训练得分：

Model saved to ./output/best_model.pth
New best model saved with average reward: 9.6
Episode:   0 | Train Reward:  25.0 | Epsilon: 0.995 | Best Eval Avg: 9.6
Episode:   1 | Train Reward:  14.0 | Epsilon: 0.990 | Best Eval Avg: 9.6
Episode:   2 | Train Reward:  12.0 | Epsilon: 0.985 | Best Eval Avg: 9.6
Episode:   3 | Train Reward:  20.0 | Epsilon: 0.980 | Best Eval Avg: 9.6
Episode:   4 | Train Reward:  40.0 | Epsilon: 0.975 | Best Eval Avg: 9.6
Episode:   5 | Train Reward:  12.0 | Epsilon: 0.970 | Best Eval Avg: 9.6
Episode:   6 | Train Reward:  22.0 | Epsilon: 0.966 | Best Eval Avg: 9.6
Episode:   7 | Train Reward:  13.0 | Epsilon: 0.961 | Best Eval Avg: 9.6
Episode:   8 | Train Reward:  15.0 | Epsilon: 0.956 | Best Eval Avg: 9.6
Episode:   9 | Train Reward:  16.0 | Epsilon: 0.951 | Best Eval Avg: 9.6
……
Model saved to ./output/best_model.pth
New best model saved with average reward: 251.6
Episode:  90 | Train Reward:  25.0 | Epsilon: 0.634 | Best Eval Avg: 251.6
Episode:  91 | Train Reward: 150.0 | Epsilon: 0.631 | Best Eval Avg: 251.6
Episode:  92 | Train Reward:  44.0 | Epsilon: 0.627 | Best Eval Avg: 251.6
Episode:  93 | Train Reward:  23.0 | Epsilon: 0.624 | Best Eval Avg: 251.6
Episode:  94 | Train Reward:  27.0 | Epsilon: 0.621 | Best Eval Avg: 251.6
Episode:  95 | Train Reward:  79.0 | Epsilon: 0.618 | Best Eval Avg: 251.6
Episode:  96 | Train Reward: 103.0 | Epsilon: 0.615 | Best Eval Avg: 251.6
Episode:  97 | Train Reward:  45.0 | Epsilon: 0.612 | Best Eval Avg: 251.6
Episode:  98 | Train Reward:  33.0 | Epsilon: 0.609 | Best Eval Avg: 251.6
Episode:  99 | Train Reward:  33.0 | Epsilon: 0.606 | Best Eval Avg: 251.6
Test Episode: 0, Reward: 228.0
Test Episode: 1, Reward: 222.0
Test Episode: 2, Reward: 207.0