search

Tencent-RoboticsX / lifelike-agility-and-play

Project Page for Lifelike Agility and Play in Quadrupedal Robots using Reinforcement Learning and Generative Pre-trained Models
Other
119 stars 11 forks source link

it seems that I got onlt 1 actor with a terminal #6

Closed Huisouan closed 4 days ago

Huisouan commented 2 weeks ago

When using TLeague, each terminal can open only one actor. Is there a way to run multiple actors simultaneously on one computer for training, similar to Isaac Gym?

Huisouan commented 2 weeks ago

image by the way,if I adjust the batch_size in actor_policy_config,ill get this valueerror

slivermoda commented 1 week ago

When using TLeague, each terminal can open only one actor. Is there a way to run multiple actors simultaneously on one computer for training, similar to Isaac Gym?

TLeague is a distributed RL infrastructure, and you can simply write a bash scripts to loop and launch multiple actors to fit your computing resources. Everything goes in a distributed way. Isaac Gym support GPU inference for environment dynamics, so training with Isaac does not need to use distributed learner-actor infra. But pybullet or other physics engine can not.