UnityActionException: behavior 3DBall?team=1 needs a continuous input of dimension (0, 2)

[nagybalint25]

Running a version of 3DBall, where I've deleted all agents except for one. Whenever I issue an action using Gym (with the wrapper), for example taking a random action using

env.action_space.sample()

I get an error:

The behavior 3DBall?team=1 needs a continuous input of dimension (0, 2) for (<number of agents>, <action size>) but received input of dimension (1, 2)

There clearly is an agent in my executable, and mlagents-learn works fine too.

My code is as follows:

from mlagents_envs.environment import UnityEnvironment
from gym_unity.envs import UnityToGymWrapper
gym.logger.set_level(40)

def main():
    unity_env = UnityEnvironment(file_name="3dballv2")
    env = UnityToGymWrapper(unity_env,  uint8_visual=True,allow_multiple_obs=True)
    env.reset()
    for _ in range(1000):
        env.render()
        env.step(env.action_space.sample()) #random action
    env.close()    
    


if __name__ == '__main__':
    main()

Environment:

• Unity Version: Unity 2020.1f1
• OS + version: Windows 10
• ML-Agents version: Release 15
• Torch version: 1.7.1
• Environment: 3DBalls

[vincentpierre]

Hi @nagybalint25
I was able to reproduce the issue. The problem is that your agent is done, do the environment must reset :
The following code should do the trick :

from mlagents_envs.environment import UnityEnvironment
from gym_unity.envs import UnityToGymWrapper
gym.logger.set_level(40)

def main():
    unity_env = UnityEnvironment(file_name="3dballv2")
    env = UnityToGymWrapper(unity_env,  uint8_visual=True,allow_multiple_obs=True)
    env.reset()
    for _ in range(1000):
        env.render()
        o,r,d,_ = env.step(env.action_space.sample()) #random action
        if d:
            env.reset()
    env.close()    
    


if __name__ == '__main__':
    main()

The error message is wrong and I will work to resolve this. Thank you for raising this issue.

[nagybalint25]

Thank you very much for your kind and quick reply! Absolutely solved my problem. Vincent-Pierre BERGES ***@***.***> ezt írta (időpont: 2021. márc. 31., Sze, 20:15):

…

Hi @nagybalint25 <https://github.com/nagybalint25> I was able to reproduce the issue. The problem is that your agent is done, do the environment must reset : The following code should do the trick : from mlagents_envs.environment import UnityEnvironment from gym_unity.envs import UnityToGymWrapper gym.logger.set_level(40) def main(): unity_env = UnityEnvironment(file_name="3dballv2") env = UnityToGymWrapper(unity_env, uint8_visual=True,allow_multiple_obs=True) env.reset() for _ in range(1000): env.render() o,r,d,_ = env.step(env.action_space.sample()) #random action if d: env.reset() env.close() if __name__ == '__main__': main() The error message is wrong and I will work to resolve this. Thank you for raising this issue. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#5204 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ASZG3QM26W56G2EUGOYKR5LTGNRD7ANCNFSM42E2OF7Q> .

[dongruoping]

Thanks @vincentpierre !

Since the fix has been merged I'll close it.

[github-actions]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.