search

DLR-RM / rl-trained-agents

A collection of pre-trained RL agents using Stable Baselines3
https://huggingface.co/sb3
MIT License
105 stars 25 forks source link

HER not compatible with zoo/stable-baselines3 #19

Closed sweber1 closed 2 years ago

sweber1 commented 2 years ago

I was trying to use the trained HER agents for Fetch tasks. But the 'her' implementation has changed in stable-baselines3 in PR 243 (https://github.com/DLR-RM/stable-baselines3/pull/243). It's no longer an algo, but instead just a type of replay buffer.

python3 enjoy.py --algo her --env FetchSlide-v1 doesn't work anymore.

I'm raising this issue to avoid other users wasting time until the configs are updated. My workaround was to downgrade stable-baselines3 and rl-baselines3-zoo to v1.0.0.

araffin commented 2 years ago

Hello,

My workaround was to downgrade stable-baselines3 and rl-baselines3-zoo to v1.0.0.

you don't need that.

we kept the old policies, but for the new ones, you can find them in the TQC folder, for instance:

so now you need to do:

python3 enjoy.py --algo tqc --env FetchSlide-v1

They are all listed in our benchmark file: