You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
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.
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.
The text was updated successfully, but these errors were encountered:
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.
The text was updated successfully, but these errors were encountered: