Resolved Unity Timeout Exception when using Heuristic Only

I am trying to train an imitation learning agent. I have entered the Heuristic function in my AI so I can control it. But if I select 'Heuristic Only' as behaviour type, the environment crashes after about 50 seconds saying Unity Timeout Exception. Full error message is below. I have Python 3.8.5, Unity 2020.3.0f1. I have tried a different Neural Net. Behaviour Type 'Default' works as expected but I can't train my agent that way. Anyone know how I could get it working again? Thanks.

Also, I'm using Anaconda3 for the virtual environment. A demo file is still created at the end of the run.

Code (csharp):

1.  
2. (mlagents-r8) E:\Programming\Unity Hub\My projects\Rocket Ball wAI>mlagents-lear
3. n --time-scale 0.002 --target-frame-rate 60 --run-id P2AIAtt5
4.  
5. (Unity logo)
6.  
7. Version information:
8.   ml-agents: 0.25.0,
9.   ml-agents-envs: 0.25.0,
10.   Communicator API: 1.5.0,
11.   PyTorch: 1.8.0
12. 2021-03-25 22:56:04 INFO [learn.py:245] run_seed set to 7283
13. 2021-03-25 22:56:04 INFO [torch.py:58] default Torch device: cuda
14. 2021-03-25 22:56:06 INFO [environment.py:210] Listening on port 5004. Start trai
15. ning by pressing the Play button in the Unity Editor.
16. 2021-03-25 22:56:12 INFO [environment.py:112] Connected to Unity environment wit
17. h package version 1.9.0-preview and communication version 1.5.0
18. 2021-03-25 22:57:18 INFO [subprocess_env_manager.py:220] UnityEnvironment worker
19. 0: environment stopping.
20. 2021-03-25 22:57:18 INFO [trainer_controller.py:81] Saved Model
21. Traceback (most recent call last):
22.   File "e:\programming\anaconda3\envs\mlagents-r8\lib\runpy.py", line 194, in _r
23. un_module_as_main
24.     return _run_code(code, main_globals, None,
25.   File "e:\programming\anaconda3\envs\mlagents-r8\lib\runpy.py", line 87, in _ru
26. n_code
27.     exec(code, run_globals)
28.   File "E:\Programming\Anaconda3\envs\mlagents-r8\Scripts\mlagents-learn.exe\__m
29. ain__.py", line 7, in <module>
30.   File "e:\programming\anaconda3\envs\mlagents-r8\lib\site-packages\mlagents\tra
31. iners\learn.py", line 250, in main
32.     run_cli(parse_command_line())
33.   File "e:\programming\anaconda3\envs\mlagents-r8\lib\site-packages\mlagents\tra
34. iners\learn.py", line 246, in run_cli
35.     run_training(run_seed, options)
36.   File "e:\programming\anaconda3\envs\mlagents-r8\lib\site-packages\mlagents\tra
37. iners\learn.py", line 125, in run_training
38.     tc.start_learning(env_manager)
39.   File "e:\programming\anaconda3\envs\mlagents-r8\lib\site-packages\mlagents_env
40. s\timers.py", line 305, in wrapped
41.     return func(*args, **kwargs)
42.   File "e:\programming\anaconda3\envs\mlagents-r8\lib\site-packages\mlagents\tra
43. iners\trainer_controller.py", line 173, in start_learning
44.     self._reset_env(env_manager)
45.   File "e:\programming\anaconda3\envs\mlagents-r8\lib\site-packages\mlagents_env
46. s\timers.py", line 305, in wrapped
47.     return func(*args, **kwargs)
48.   File "e:\programming\anaconda3\envs\mlagents-r8\lib\site-packages\mlagents\tra
49. iners\trainer_controller.py", line 105, in _reset_env
50.     env_manager.reset(config=new_config)
51.   File "e:\programming\anaconda3\envs\mlagents-r8\lib\site-packages\mlagents\tra
52. iners\env_manager.py", line 68, in reset
53.     self.first_step_infos = self._reset_env(config)
54.   File "e:\programming\anaconda3\envs\mlagents-r8\lib\site-packages\mlagents\tra
55. iners\subprocess_env_manager.py", line 333, in _reset_env
56.     ew.previous_step = EnvironmentStep(ew.recv().payload, ew.worker_id, {}, {})
57.   File "e:\programming\anaconda3\envs\mlagents-r8\lib\site-packages\mlagents\tra
58. iners\subprocess_env_manager.py", line 98, in recv
59.     raise env_exception
60. mlagents_envs.exception.UnityTimeOutException: The Unity environment took too lo
61. ng to respond. Make sure that :
62.          The environment does not need user interaction to launch
63.          The Agents' Behavior Parameters > Behavior Type is set to "Default"
64.         The environment and the Python interface have compatible versions.
65.  

 

Ok, so you don't connect to the port when you're training an imitation learning agent, you just record the demo file. Egg is on my face.