Hi. I have a very simple script, one cube agent, one capsule target, on a plane floor with some cube walls it has to navigate around. I’ve been able to train the environment using the mlagents-learn command on my local device with the built-in PPO option no problem. Everything is fine, training great. I tested inference, that worked too.
Okay now I want to connect to the ml-envs python API so I can start using my own learning algorithm. As per the official docs and Colab tutorial, I used the following code to load the environment from my built binary file:
UnityTimeOutException: The Unity environment took too long to respond. Make sure that :
The environment does not need user interaction to launch
The Agents' Behavior Parameters > Behavior Type is set to "Default"
The environment and the Python interface have compatible versions.
I do not have any code (not even heuristic) that requires user input.
My agents behavior is set to default and no neural net is loaded into the behavior parameters model.
On my local, where I built the environment, my specs are:
Ubuntu 18.04.5 LTS os
Python 3.6.9
mlagents==0.20.0
Where I am trying to connect to the ml-envs API, my specs are:
Google Colaboratory “os”
Python 3.6.9
mlagents==0.20.0
In Colab, I put the binary x86_64 file in the upload file section.
What is the problem? I also considered building the environment in “headless” mode but I need camera observation, so I don’t think that will work. I also made sure when I built my environment, I selected the target platform as Linux, x86_64, and selected my scene.
Totally lost on how to recover, any help is appreciated! I attached a file of my agent’s script and important component settings.
Update:
So I can reproduce this error in my local python environment on my host machine (which I was previously training the unity using mlagents-learn part of the api). But, the interesting thing is that after some banging my head against my computer I noticed the environment seems to be loaded even though it is throwing the error. So potential bug?
Edit:
I should add more details here. When I call UnityEnvironment(file_name=“file”, seed=1), it tries to open the build file and start playing the scene but crashes early on. Then if I try to access anything about the env I get the error above or null return.
Update 2:
I reproduced the same error using the 3DBall example. So the issue is not how I made my environment. My python API is 1.1.0 and the communicator API is 1.0.0. But this shouldn’t be an issue right since the main version is same? I also tried using unity v.2019.4.11f1 and v.2020.1.7f1. I’m using the ml-agents package version 1.5.0 in the environment but also tried switching that with same error persisting.
Hi, sorry for the late reply. Those versions should be compatible and the Display Resolution Dialogue is removed in newer Unity releases so I wouldn’t worry about it if it doesn’t show up in your settings.
If you use the mlagents-learn command with --env=<your_executable> does it connect as expected or was your earlier comment referring to the in-editor training?
Hi! Thank you. Yes, it mlagents-learn works fine. I want to use my own learning algorithm though with the python API. So I don’t want to use mlagents-learn right? I need to use mlagents_envs. This is my understanding, is it correct?
Thanks for the info. If you can run the same executable via mlagents-learn then the executable should be built in a way that works with the python API and your build settings should be correct. I will try to reproduce this with 3DBall and the listed versions.
To make sure this isn’t a Colaboratory issue can you try running directly from the command line in a python3 interpreter in the directory where you have the 3DBall executable?
python3
```python
from mlagents_envs.environment import UnityEnvironment
env = UnityEnvironment(file_name="3DBall", seed=1, side_channels=[])
env.reset()
for i in range(1000):
env.step()
exit()
Hi again, thank you very much for confirming this. I did run it in python terminal directly and received the same error so it is not a Colab dependent issue. I want to add that when I run the code you provided, after I call UnityEnvirontment it tries to open the game, crashes upon loading, and either produces the error in the terminal or gets stuck on the function itself.
Hmm, the fact that this is failing for 3DBall is concerning. Before I try to emulate your setup can you confirm the version of your Unity package in-editor (window → Package Manager → ML Agents).
along with the versions of mlagents and the version of mlagents-envs?
Are you building the 3dBall example with headless mode? If not can you please try that. If 3dBall still fails then this is some bug we will try to file on github.
Hi, I was following this thread and I am getting a similar problem. I have built my exe file and I was able to execute using python. But, after running for sometime it says:
INFO:mlagents_envs:Environment timed out shutting down. Killing…
mlagents_envs.exception.UnityTimeOutException: The Unity environment took too long to respond. Make sure that :
The environment does not need user interaction to launch
The Agents are linked to the appropriate Brains
The environment and the Python interface have compatible versions.
I’m using Python Version:
Python 3.9.0 (tags/v3.9.0:9cf6752, Oct 5 2020, 15:34:40) [MSC v.1927 64 bit (AMD64)] on win32
Error Message:
mlagents_envs.exception.UnityTimeOutException: The Unity environment took too long to respond. Make sure that :
The environment does not need user interaction to launch
The Agents’ Behavior Parameters > Behavior Type is set to “Default”
The environment and the Python interface have compatible versions.
If you’re running on a headless server without graphics support, turn off display by either passing --no-graphics option or build your Unity executable as server build.
I don’t seem to find any solution for this PLEASE HELP!
Also, the Machine Learning module runs and the script can work however the module do not safe because of the UnityTimeOutException.
