• I have marked all applicable categories:
  • exception-raising bug
  • RL algorithm bug
  • documentation request (i.e. "X is missing from the documentation.")
  • new feature request
• I have visited the source website, and in particular read the known issues
• I have searched through the issue tracker and issue categories for duplicates
• I have mentioned version numbers, operating system and environment, where applicable:

  import tianshou, torch, sys
  print(tianshou.__version__, torch.__version__, sys.version, sys.platform)

hello~
In most envs, env step cost almost the same time, so SubprocVectorEnv and RayVectorEnv run almost linear scale up. However in my env， each step cost very different time， e.g. 90% step cost 1s, but 10% cost 10s. So when I run 10 SubprocVectorEnv，there are almost no speeding up, because almose envery step, there is a slow step cost 10s, and other 9 fast step cost 1s, but the step 'blocked' to wait for all 10 envs step finished. I think if the 10 envs step independently, the total time can almost linear scale up. But it seems that I have to change collector too much. Do you have any suggestion? Thanks very much for you great work on tianshou~