Understanding Mujoco Humanoid Standup
If you are looking for information about Mujoco Humanoid Standup, you have come to the right place. Reward 258k.
Key Takeaways about Mujoco Humanoid Standup
- Even though it hit 300k, it doesn't
- Took 5785000 frames to train and stands up and moves much better than C-TD3 with similar reward.
- The video was created in a public tutorial colab notebook. Follow this link to try it yourself: ...
- Similar to previous video (https://www.youtube.com/watch?v=DCJbQXCaBAA) but now attempting to learn to recover from the ...
- PPOC based option critic learner with safety added as regularized variance in return.
Detailed Analysis of Mujoco Humanoid Standup
Similar to previous video (https://www.youtube.com/watch?v=OK6Epi-QL9Y) but with position-controlled joints for the Test reward 254k, recorded 255k. Test reward 227k, during recording 230k It started learning at frame 4.9M and unfortunately I had a cut of at 5M.
This environment is described in this paper: https://arxiv.org/abs/2006.12983 The agent used is Abdolmaleki's MPO (2018): ...
We hope this detailed breakdown of Mujoco Humanoid Standup was helpful.