search

fan-ziqi / rl_sar

Simulation verification and physical deployment of robot reinforcement learning algorithms, suitable for quadruped robots, wheeled robots, and humanoid robots. "sar" stands for "simulation and real"
Apache License 2.0
90 stars 7 forks source link

How to choose data for Cyberdog training? #4

Closed RubSevian closed 1 month ago

RubSevian commented 1 month ago

I want to repeat your success and possibly improve walking, and for this I really need information about where you conducted and what you based your Cyberdog training on.Please

fan-ziqi commented 1 month ago

I use legged_gym as the base framework, and the model is based on a modified version of cyberdog1_description with parameters provided by the official repository cyberdog1_description. The environment setup uses the script rl_docker.

However, after testing, it was found that using Kp=20 and Kd=0.5 does not enable walking on the real dog. Therefore, it is recommended to use the actuator_network instead of PD for training.

RubSevian commented 1 month ago

Hello, I have assembled your rl_docker.I would like to learn more about the Cyberdog robot training process, can you tell me more about it?

fan-ziqi commented 1 month ago

Please join Discord group in README.md