The Isaac Sim documentation for reinforcement learning suggests that I should be able to run
python.sh train.sh out of the box, however if you do this in the Isaac Sim docker container, you get
root@ip-172-31-78-124:/isaac-sim# ./python.sh standalone_examples/api/omni.isaac.jetbot/stable_baselines_example/train.py running as root Omniverse System Monitor already running. Traceback (most recent call last): File "standalone_examples/api/omni.isaac.jetbot/stable_baselines_example/train.py", line 10, in <module> from env import JetBotEnv File "/isaac-sim/standalone_examples/api/omni.isaac.jetbot/stable_baselines_example/env.py", line 9, in <module> import gym ModuleNotFoundError: No module named 'gym' There was an error running python
This documentation link should reference the python environment documentation:
This documentation also has issues,
./python.sh -m pip install gym results in the error
no such option: --allow-root.
Anyways, removing the offending line from
args="$args --allow-root") fixes the pip issue.
Too many layers of redirection can cause issues like this. Ideally I’d just run
python -m pip install gym and
python train.py, and docker or anaconda would load the python environment. Does the Isaac team have any plans to move Isaac Sim / Gym to NVIDIA Corporation · GitHub ? There’s a few issues like this that I’d rather submit a PR for than write a bug report on a discussion forum.