Greetings.
I was quite inspired by your work on theory of mind and it;s integration with AI agents.
Thank you very much for publishing the experiments, as well as the experiment simulator.
Regarding the latter, I was hoping you could provide a bit more details regarding the action space for the overcooked environment featured in this repository.
I went as far as looking into the OvercookedEnvironment class's step method, but it is not clear what I should feed when calling env.step(action_dict)
.
The paper mentions an action space size 5 to move the agent: North
, East
, West
, South
, Stay still
.
I tried passing action_dict={"agent-1": 0,1,2,3,4... etc...}
but it did not seem to work consistently.
For a bit more context, I am trying to use this environment with off-the-shelf RL algorithms such as those provided in Stable Baselines 3, so it would be really helpful to have a specification for env.action_space
.
Looking forward to hear from you.
Best regards.