arnaudfickinger / gym-multigrid Goto Github PK

Hey! Thank you for making this extension to minigrid. Do you plan to add more multi-agent environments in the future such as pursuit evasion, warehouse environment or more environments from literature?

Env.render() returns the observation of each agent in the environment. This observation is used to choose the best policy in q-learing. However the env.render() returns an array of arrays containing 2-dim arrays with values I do not understand. Is there anyone who can clarify this to me? How could these observations be used to conduct q-learning?

I am trying to train the agents in the multigrid environment using the torch-rl repository (https://github.com/lcswillems/rl-starter-files) which is compatible with gym-minigrid. After registering the new environment in gym (multigrid-collect-v0) and running the environment using

python3 -m scripts.train --algo ppo --env multigrid-collect-v0

I am receiving the following runtime error: RuntimeError: Expected 4-dimensional input for 4-dimensional weight [16, 3, 2, 2], but got 5-dimensional input of size [16, 3, 5, 3, 6] instead

Is there a way to make it compatible with the torch-rl repository or am I making a mistake somewhere? Any help would be much appreciated, thanks!