search

fangxiaoshen / ROS_pytorch_RL

在turtlebot3，pytorch上使用DQN,DDPG,PPO,SAC算法，在gazebo上实现仿真。Use DQN, DDPG, PPO, SAC algorithm on turtlebot3, pytorch on turtlebot3, pytorch, and realize simulation on gazebo. Use DQN, DDPG, PPO, SAC algorithm to realize simulation on gazebo.
Apache License 2.0
87 stars 10 forks source link

readme

ROS_pytorch_RL

在turtlebot3，pytorch上使用DQN,DDPG,PPO,SAC算法，在gazebo上实现仿真。

仿真配置教程

https://blog.csdn.net/qq2650326396/article/details/124801005

实车配置教程

https://blog.csdn.net/qq2650326396/article/details/132076450

我设置了新的reward和控制，让小车减少碰撞。在PPO下训练了140回合后的训练效果：

可以看到小车一定不会碰撞，因为我加了一个膨胀的距离给小车。 [INFO] [1639030326.511617, 504.920000]: Ep: 20 score: 9586.48 memory: 52521 episode_step: 2500.00 time: 2:57:02

[INFO] [1639030395.362659, 68.758000]: Goal position : 0.4, 0.4

[INFO] [1639030497.016797, 170.313000]: Goal position : 6.5, 4.5

[INFO] [1639030571.016659, 244.231000]: Goal position : 0.4, 0.4

[INFO] [1639030682.929499, 356.027000]: Goal position : 6.5, 4.5

[INFO] [1639030761.944676, 434.946000]: Goal position : 0.4, 0.4

[INFO] [1639030830.140551, 503.066000]: Goal position : 6.5, 4.5 [INFO] [1639030830.147300, 503.072000]: time out! [INFO] [1639030831.429489, 504.350000]: Ep: 21 score: 7548.49 memory: 55022 episode_step: 2500.00 time: 3:05:27