search

Alescontrela / AMP_for_hardware

Code for "Adversarial Motion Priors Make Good Substitutes for Complex Reward Functions"
Other
163 stars 41 forks source link

How to balance task_reward and style_reward #21

Open hanzhi0410 opened 1 month ago

hanzhi0410 commented 1 month ago

I want to use amp training to walk on complex terrain. Have you tried it before? What parameters do you think have a significant impact on terrain adaptability

yinkangning0124 commented 1 month ago

I want to use amp training to walk on complex terrain. Have you tried it before? What parameters do you think have a significant impact on terrain adaptability

Do you have any ideas? I am also trying to train walking//running on terrain. I wonder the composition of the task reward and the balance between task reward and style reward.

hanzhi0410 commented 1 month ago

I want to use amp training to walk on complex terrain. Have you tried it before? What parameters do you think have a significant impact on terrain adaptability

Do you have any ideas? I am also trying to train walking//running on terrain. I wonder the composition of the task reward and the balance between task reward and style reward.

This balance is difficult to quantify, and I believe it is related to many factors. Therefore, during training, I continuously increased the style reward from scratch, and made choices through multiple experimental results. I also found that excessive style rewards can affect the robot's terrain robustness and speed following tasks .