search

edbeeching / godot_rl_agents

An Open Source package that allows video game creators, AI researchers and hobbyists the opportunity to learn complex behaviors for their Non Player Characters or agents
MIT License
942 stars 69 forks source link

Training an agent with WSL and sample-factory is not learning #89

Closed visuallization closed 1 year ago

visuallization commented 1 year ago

Training an agent with WSL and sample-factory is not learning.

When running on Linux: drl --trainer=sf --env=gdrl --env_path=examples/godot_rl_JumperHard/bin/JumperHard.x86_64 --num_workers=10 --experiment=JumperHard01 --viz --batched_sampling=True --speedup=8

The training looks something like this:

image

When running on WSL with the same command the training looks like this:

image-2

As you can see on Linux the agent actually learns, whereas on WSL the agent does not learn.

This might be related to this issue: https://github.com/alex-petrenko/sample-factory/issues/128

As stated in the issue you might be able to work around this by running the training on the CPU but it won't allow for large scale training.

edbeeching commented 1 year ago

Closing for now, but keeping pinned.