Gym Wrapper produces Tuple instead of Dict for multiple observations (does not work with SB3)

[leo2r]

Describe the bug I've used Unity and ml-agents to create a custom environment. I then built the exe file (binary file) to use the UnityToGymWrapper to create a Gym instance of the environment. I'm now trying to use stable-baselines3 to train my agent using 'MultiInputPolicy', since the observations are (image1, image2, image3, vector(ie 'direct' features)). However, I get the following error:

raise NotImplementedError(f"{observation_space} observation space is not supported")

When comparing the returned observation from my UnityToGymWrapper env and an example env, I spot this difference:

UnityToGymWrapper env obs:

Tuple(Box([[[0] [0] [0] ... [255] [255] [255]]], (84, 84, 1), uint8), Box([[[0] [0] [0] ... [255] [255] [255]]], (84, 84, 1), uint8), Box([[[0] [0] [0] ... [255] [255] [255]]], (84, 84, 1), uint8), Box([-inf ... -inf], [inf ... inf], (13,), float32))

SB3 env obs:

{'vec': array([0.72354802, 0.18255346, 0.62559858, 0.69045166, 0.03119092]), 'img': array([[[ 6], [222], [204], ..., [237], [182], [209]]], dtype=uint8)}

So it seems like the UnityToGymWrapper env is returning a Tuple as multi observations whereas SB3 (and therefore Gym) want/need a dictionary. Is this the desired way to present multi observations? Seems like the Gym wrapper should create obs in the Gym format? How can I create a work around for this? Thank you :)

To Reproduce Steps to reproduce the behavior:

1. env_exe =
2. unity_env = UnityEnvironment(env_exe)
3. env = UnityToGymWrapper(unity_env, uint8_visual=True, allow_multiple_obs=True)3.
4. from stable_baselines3 import PPO
5. model = PPO('MultiInputPolicy', gym_env)6.
6. Error happens

Environment (please complete the following information):

• Unity Version: Unity 2021.3.5.f1
• OS + version: Ubuntu 20.04.4 LTS
• ML-Agents version: v0.27.0
• Torch version: 1.8.1
• Environment: Custom

[hanlanyi]

Is that true that all people work for this project in Unity were fired?

[stale[bot]]

This issue has been automatically marked as stale because it has not had activity in the last 28 days. It will be closed in the next 14 days if no further activity occurs. Thank you for your contributions.

[miguelalonsojr]

Is that true that all people work for this project in Unity were fired?

No. There have been changes at Unity that have slowed our development substantially, but we're still alive! :) Thanks for reaching out. I'll have a look at this.

[stale[bot]]

This issue has been automatically marked as stale because it has not had activity in the last 90 days. It will be closed in the next 30 days if no further activity occurs. Thank you for your contributions.

[aha85b]

I am having this issue, any one got it fixed?

[zbwby819]

Same issue, any updates?