facebookresearch / ScaDiver

Project for the paper "A Scalable Approach to Control Diverse Behaviors for Physically Simulated Characters"
Other
152 stars 18 forks source link

Questions about training. #8

Open ZC119 opened 3 years ago

ZC119 commented 3 years ago

Hi, I trained on the single test motion clip data/motion/test.bvh with default configuration in test_env_humanoid_imitation.yaml. The total reward does not increase after 12 hours training. However, the character still fall. I wonder if there are some parameters in configuration need to be tuned.

commond: python rllib_driver.py --mode train --spec data/spec/test_env_humanoid_imitation.yaml --project_dir ./

reward: Screenshot 2021-05-27 202821

test results: 2

Thanks !