Skip to content

unitree Go2 robot learns locomotion with N-P3O algorithm and HIM alike policy trained by isaacgym

Notifications You must be signed in to change notification settings

xyyl/LocomotionWithNP3O

 
 

Repository files navigation

The performance depoly on real robot please checkout link below:

【感觉比之前好多了呢(rl运控实验成功系列)欢迎大家给我的repo加star~-哔哩哔哩】 https://b23.tv/LyjHWJG

打扰大家一下,再给我的repo骗骗star,视频里展示的方案是类似himloco的单阶段训练方案,一样用到了对比学习,不过与himloco的swav不一样的是我用了barlow twin这个算法,它是没有prototype的概念的,同时我的对比学习目标不是拉近history和未来t+1的latent的相似度而是拉近t:t-5 与 t+1:t-4之间的差异(公用mlp encoder与对比学习论文的结构设计更是吻合),我的方案2000轮能收敛到terrain level 6,目前看着还行,据我所知应该没人有这么做过的,由于我是业余爱好,所以我也没发论文的时间,大家有兴趣的话可以看看我的代码一起讨论进步下

About

unitree Go2 robot learns locomotion with N-P3O algorithm and HIM alike policy trained by isaacgym

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%