The reset process for each episode is very very slow.

google-deepmind / dm_control

Google DeepMind's software stack for physics-based simulation and Reinforcement Learning environments, using MuJoCo.

Apache License 2.0

3.72k stars 660 forks source link

The reset process for each episode is very very slow. #272

Closed Frank-Dz closed 2 years ago

Frank-Dz commented 2 years ago

I am using the soccer 2-vs-2 for training. I found the self.timestep = self.env.reset() process takes a long time after each epsiode. Any suggestion about that?

Frank-Dz commented 2 years ago

A similar problem is posted here: https://github.com/deepmind/dm_control/issues/132

I don't know what do they mean by "simulation" is slow. For me, the simulation part (time.step()) is fast enough, but the machine almost gets stuck there when I try to reset.

I have 16 threads and 2 teams (4 agents) here. I am using Box robot to play football.

AwesomeLemon commented 1 year ago

Hi @Frank-Dz, were you able to figure out the solution to this issue?