Exploring Trpo Mujoco
Exploring Trpo Mujoco reveals several interesting facts.
- Hopper C-R-TD3-500 Reward during training 3803, during recording 3795.
- Mujoco
- Test reward 227k, during recording 230k It started learning at frame 4.9M and unfortunately I had a cut of at 5M.
- Reward 258k.
- Took 5785000 frames to train and stands up and moves much better than C-TD3 with similar reward.
In-Depth Information on Trpo Mujoco
TRPO MuJoCo MuJoCo Even though it hit 300k, it doesn't stand up Strangely C-RA-TD3-500 with reward 283k does stand up. Neural Network trained by the proposed hardware accelerated training scheme (hardware accelerated Lightweight Simulator + ...
For more details see publication [1] and full-length video [2]. [1] https://arxiv.org/abs/2105.12196 [2] ...
Stay tuned for more updates related to Trpo Mujoco.
