JaxNav, base class update, version increment, scipy bound

FLAIROx / JaxMARL

Multi-Agent Reinforcement Learning with JAX

Apache License 2.0

393 stars 68 forks source link

JaxNav, base class update, version increment, scipy bound #99

Closed amacrutherford closed 2 months ago

amacrutherford commented 2 months ago

Couple of things:

JaxNav environment (gif of this replaces switch riddle in the readme collage)
Update to step to allow for autoresets to a specified state, this should not change any current functionality and removes need for a wrapper when doing UED.
Added get_avail_actions to the base class
scipy bounded to =<1.12 as i experienced a bug with 1.13.1
also changed the order of installing tmux and pip install -e . in the Dockerfile as this should make rebuilds marginally quicker

because of the change to step, have incremented the version counter

benellis3 commented 2 months ago

Have we checked the other environments run with that version of SciPy? If not would be good to check that before merging

amacrutherford commented 2 months ago

this is a good point, I have added random action rollout tests for the other environments (i.e. rollout 10 steps to make sure no errors are thrown). Will also do this for the learning scripts, maybe train for like 2000 environment timesteps or smth just to check everything runs

mttga commented 2 months ago

do we want to include the tests for the learning scripts? Or are we happy with the new environment tests (which look good) for this PR?

amacrutherford commented 2 months ago

do we want to include the tests for the learning scripts? Or are we happy with the new environment tests (which look good) for this PR?

have now added :slightly_smiling_face:, let me know what ya think