Multiple Unity instances stops

[ProxyCausal]

Unity environment worker stops almost immediately after running when using num-envs > 1.

Note:
Slightly unrelated issue, but I can't make a separate post about it since it only happens on my custom environment. When I created multiple training areas and start training, my agent won't move after pressing play. I am using a separate controller script instead of heuristics mode, and I can manually control with that. In my original scene with only one copy, the opposite is true: I can train but not manually control. When I press play the position moves, but it is not moving in editor and I can't control. Not sure why this is, but I suspect somehow it is still being controlled by ml-agents.

To Reproduce
Steps to reproduce the behavior:

1. CD into ml-agents directory, run mlagents-learn config/ppo/WallJump.yaml --run-id=WallJumper --num-envs=2

Console logs / stack traces

WARNING:tensorflow:From c:\users\gdev\anaconda3\envs\rl\lib\site-packages\tensorflow_core\python\compat\v2_compat.py:65: disable_resource_variables (from tensorflow.python.ops.variable_scope) is deprecated and will be removed in a future version.
Instructions for updating:
non-resource variables are not supported in the long term
WARNING:tensorflow:From c:\users\gdev\anaconda3\envs\rl\lib\site-packages\tensorflow_core\python\compat\v2_compat.py:65: disable_resource_variables (from tensorflow.python.ops.variable_scope) is deprecated and will be removed in a future version.
Instructions for updating:
non-resource variables are not supported in the long term
2020-07-03 13:14:48 INFO [subprocess_env_manager.py:191] UnityEnvironment worker 1: environment stopping.
2020-07-03 13:14:48 INFO [environment.py:199] Listening on port 5004. Start training by pressing the Play button in the Unity Editor.
2020-07-03 13:14:58 INFO [environment.py:108] Connected to Unity environment with package version 1.1.0-preview and communication version 1.0.0
2020-07-03 13:14:58 INFO [environment.py:265] Connected new brain:
SmallWallJump?team=0
2020-07-03 13:14:58 INFO [trainer_controller.py:234] Learning was interrupted. Please wait while the graph is generated.
2020-07-03 13:14:58 INFO [trainer_controller.py:108] Saved Model
Traceback (most recent call last):
  File "c:\users\gdev\anaconda3\envs\rl\lib\runpy.py", line 193, in _run_module_as_main
    "__main__", mod_spec)
  File "c:\users\gdev\anaconda3\envs\rl\lib\runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "C:\Users\gdev\anaconda3\envs\rl\Scripts\mlagents-learn.exe\__main__.py", line 7, in <module>
  File "c:\users\gdev\anaconda3\envs\rl\lib\site-packages\mlagents\trainers\learn.py", line 322, in main
    run_cli(parse_command_line())
  File "c:\users\gdev\anaconda3\envs\rl\lib\site-packages\mlagents\trainers\learn.py", line 318, in run_cli
    run_training(run_seed, options)
  File "c:\users\gdev\anaconda3\envs\rl\lib\site-packages\mlagents\trainers\learn.py", line 163, in run_training
    tc.start_learning(env_manager)
  File "c:\users\gdev\anaconda3\envs\rl\lib\site-packages\mlagents_envs\timers.py", line 305, in wrapped
    return func(*args, **kwargs)
  File "c:\users\gdev\anaconda3\envs\rl\lib\site-packages\mlagents\trainers\trainer_controller.py", line 243, in start_learning
    raise ex
  File "c:\users\gdev\anaconda3\envs\rl\lib\site-packages\mlagents\trainers\trainer_controller.py", line 214, in start_learning
    self._reset_env(env_manager)
  File "c:\users\gdev\anaconda3\envs\rl\lib\site-packages\mlagents_envs\timers.py", line 305, in wrapped
    return func(*args, **kwargs)
  File "c:\users\gdev\anaconda3\envs\rl\lib\site-packages\mlagents\trainers\trainer_controller.py", line 150, in _reset_env
    env.reset(config=sampled_reset_param)
  File "c:\users\gdev\anaconda3\envs\rl\lib\site-packages\mlagents\trainers\env_manager.py", line 67, in reset
    self.first_step_infos = self._reset_env(config)
  File "c:\users\gdev\anaconda3\envs\rl\lib\site-packages\mlagents\trainers\subprocess_env_manager.py", line 295, in _reset_env
    ew.previous_step = EnvironmentStep(ew.recv().payload, ew.worker_id, {}, {})
  File "c:\users\gdev\anaconda3\envs\rl\lib\site-packages\mlagents\trainers\subprocess_env_manager.py", line 92, in recv
    raise env_exception
mlagents_envs.exception.UnityEnvironmentException: If the environment name is None, the worker-id must be 0 in order to connect with the Editor.

Environment:

• Unity Version: 2019.3.5f1
• OS + version: Windows 10
• ML-Agents version: 0.17.9
• TensorFlow version: 2.0.1
• Environment: WallJump

[chriselion]

mlagents-learn config/ppo/WallJump.yaml --run-id=WallJumper --num-envs=2

This isn't a valid combination of commandline parameters: you either need to omit --num-envs (if you want to connect to the editor) or add --env and point it to an executable.

We need to validate the arguments better and raise a clearer error message in this case.

Note that --num-envs refers to the number of instances of the executable to run, not the number of training areas in the scene.

[ProxyCausal]

We need to validate the arguments better and raise a clearer error message in this case.

Yes, it wasn't clear to me that I needed the executable from the training docs (I should've thought of it though).

Note that --num-envs refers to the number of instances of the executable to run, not the number of training areas in the scene.

I know, I only went the multiple instance way after the problems I was facing with multiple training areas as I noted in my post

[chriselion]

I logged the --num-envs bug as MLA-1145 in our internal tracker.

If you're still having problems with multiple areas, can you post it on the ML-Agents forum (https://forum.unity.com/forums/ml-agents.453/) ? That's a better venue for that type of questions.

[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.