OpenAI Gym ProcGen - Getting Action Meanings

Question

In the OpenAI ProcGen gym, I am not getting a way to get the meanings of the action values, I can see that there are 15 actions for the coinrun environment using env.action_space.n. I have tried both the Gym and the Gym3 versions.

This is how I make the environment (gym version).

env = gym.make('procgen:procgen-%s-v0'%('coinrun'))

Neither of these are working to seem to work.

env.action_spec()
env.env.get_action_meanings()

I have tried to change env with env.env and env.env.env, nothing works. I get the message: AttributeError: 'ToGymEnv' object has no attribute 'get_action_meanings'.

Please tell me how I can get the labelled action list.

Object Types: env is a ToGymEnv object, env.env is , and env.env.env is ProcgenGym3Env.

Answer 1

The action meanings are hidden in the get_combos method in the procgen.env.

def get_combos(self):
    return [
        ("LEFT", "DOWN"),
        ("LEFT",),
        ("LEFT", "UP"),
        ("DOWN",),
        (),
        ("UP",),
        ("RIGHT", "DOWN"),
        ("RIGHT",),
        ("RIGHT", "UP"),
        ("D",),
        ("A",),
        ("W",),
        ("S",),
        ("Q",),
        ("E",),
    ]

To access them from an environment created in gym (procgen==0.10.4) you can use:

env = gym.make('procgen:procgen-%s-v0'%('coinrun'))
print(env.unwrapped.env.env.combos)

Which will result in:

[('LEFT', 'DOWN'), ('LEFT',), ('LEFT', 'UP'), ('DOWN',), (), ('UP',), ('RIGHT', 'DOWN'), ('RIGHT',), ('RIGHT', 'UP'), ('D',), ('A',), ('W',), ('S',), ('Q',), ('E',)]

In order to acces them from a ProcgenEnv use:

env = procgen.ProcgenEnv(env_name='maze', num_envs=1)
print(env.env.combos)