[Bug Report] ValueError: Could not find module file for: 'isaaclab_tasks.manager_based.manipulation.pick_place.agents'

[LeoE-SJSU]

Describe the bug

Hello,

First time posting on GitHub so hopefully I can provide full context.

I am trying to run the imitation learning tutorial found here: https://isaac-sim.github.io/IsaacLab/main/source/overview/imitation-learning/teleop_imitation.html

But whenever I get to the training part of the GR1T2, i.e:

./isaaclab.sh -p scripts/imitation_learning/robomimic/train.py
--task Isaac-PickPlace-GR1T2-Abs-v0 --algo bc
--normalize_training_actions
--dataset ./datasets/generated_dataset_gr1.hdf5

I get the following error:

Traceback (most recent call last):
File "/home/leo/Desktop/IsaacLab/scripts/imitation_learning/robomimic/train.py", line 455, in
main(args)
File "/home/leo/Desktop/IsaacLab/scripts/imitation_learning/robomimic/train.py", line 372, in main
raise ValueError(f"Could not find module file for: '{mod_name}'")
ValueError: Could not find module file for: 'isaaclab_tasks.manager_based.manipulation.pick_place.agents'

Steps to reproduce

Run the training for GR1T2 robomimic after downloading the annotated demos.

System Info

• Commit:
• Isaac Sim Version: 5.1.0
• OS: Ubuntu 24.04
• GPU: RTX 5080
• CUDA: 13.0
• GPU Driver: 580.65.06

Additional context

I updated my installation (I have IsaacSim installed through PIP, and pulled a new IsaacLab).
I can run all other tutorials fine, including the Isaac-Stack-Cube-Franka training that appears higher on that imitation learning page to good results.
And I can even gather and annotate data using XR on a docker.
But I cannot train it.

Please help, thanks!

Checklist

• [ x] I have checked that there is no similar issue in the repo (required)
• [ x] I have checked that the issue is not in running Isaac Sim itself and is related to the repo

Acceptance Criteria

Add the criteria for which this task is considered done. If not known at issue creation time, you can add this once the issue is assigned.

• Criteria 1
• Criteria 2

[RandomOakForest]

Thank you for posting this. The error indicates that robomimic cannot import the GR1T2 pick‑place task’s agents module at the Python package path isaaclab_tasks.manager_based.manipulation.pick_place.agents.¹²

What is likely wrong

Most likely causes:

• The isaaclab_tasks package for your checkout is not installed (or not installed in editable mode) into the same Python environment where you run the robomimic training script.³⁴
• You pulled a newer IsaacLab checkout, but your training still uses an older or incomplete isaaclab_tasks tree that doesn’t include the GR1T2 pick‑place agents module.⁵²
• Your ./isaaclab.sh launch does not set PYTHONPATH to include the IsaacLab repo root, so Python cannot resolve isaaclab_tasks.manager_based.manipulation.pick_place.agents even though the files exist.³

The GR1T2 environment is one of the newer imitation‑learning tasks described in the Isaac Lab docs, built on top of the modular “manager‑based” environment stack. That stack relies on multiple Python modules under manager_based/manipulation/pick_place, including configuration, MDP, and agents code.²¹

Checks to run

Run all of these from the same environment and shell you use for ./isaaclab.sh:

1. Confirm the pick‑place files exist in your checkout

From your IsaacLab repo root:

cd /home/leo/Desktop/IsaacLab
find source -maxdepth 8 -path "*isaaclab_tasks/manager_based/manipulation/pick_place*" -print

You should see entries like:
- .../isaaclab_tasks/manager_based/manipulation/pick_place/__init__.py
- .../isaaclab_tasks/manager_based/manipulation/pick_place/pickplace_gr1t2_env_cfg.py
- .../isaaclab_tasks/manager_based/manipulation/pick_place/mdp.py (name may vary slightly per version)
- .../isaaclab_tasks/manager_based/manipulation/pick_place/agents.py or an agents package

If the agents.py (or equivalent agents subpackage) is missing but your docs and scripts refer to GR1T2, your local source tree is out of sync with the Isaac Lab version that includes this environment.⁵¹²

2. Install IsaacLab tasks in editable mode

Ensure the isaaclab_tasks Python package from your current checkout is installed in the environment used by Isaac Sim 5.1.0.⁴⁶³

cd /home/leo/Desktop/IsaacLab/source
pip install -e .

That makes isaaclab_tasks resolve directly from your source/ tree, which is the recommended layout for the Isaac Lab tutorials under Isaac Sim.⁴³

3. Test the Python import explicitly

From the IsaacLab repo root:

cd /home/leo/Desktop/IsaacLab
./isaaclab.sh -p -c "import isaaclab_tasks.manager_based.manipulation.pick_place.agents as agents; print(agents.__file__)"

- If this prints a valid path into your `source/isaaclab_tasks/.../pick_place` directory, then the agents module is importable and your earlier error is likely due to an outdated script or task name mismatch.
- If this fails with `ModuleNotFoundError` or a similar error, the module truly is not visible to Python (either missing file or wrong `PYTHONPATH`).[^3]

4. Always launch from the IsaacLab repo root

The isaaclab.sh helper script configures environment variables (including Python path) assuming it is run from the IsaacLab root, as shown in the official Isaac Lab tutorials embedded in the Isaac Sim docs.⁴³

Use:

cd /home/leo/Desktop/IsaacLab
./isaaclab.sh -p scripts/imitation_learning/robomimic/train.py \
  --task Isaac-PickPlace-GR1T2-Abs-v0 --algo bc \
  --normalize_training_actions \
  --dataset ./datasets/generated_dataset_gr1.hdf5

and avoid running ./isaaclab.sh from other directories or via different wrappers.

Version / branch alignment (Isaac Sim 5.1.0 + Isaac Lab 2.3.x)

NVIDIA recommends pairing specific Isaac Lab versions with specific Isaac Sim releases. For Isaac Sim 5.1.0, the associated Isaac Lab 2.3.x branch contains the new teleoperation and imitation‑learning flows, including GR1T2 and Mimic integration discussed in:⁷⁶³

• Isaac Lab overview and tutorials in the Isaac Sim documentation.³⁴
• NVIDIA blog post introducing Isaac Lab 2.3 with expanded teleoperation and imitation learning support.⁸

If you are on the main branch of IsaacLab while following documentation that corresponds to a specific 2.3.x snapshot, you can end up with:

• Config / task registry entries for Isaac-PickPlace-GR1T2-Abs-v0 but missing or renamed support modules (such as agents), or
• Slightly different module paths or filenames than those expected by scripts/imitation_learning/robomimic/train.py.¹²⁵

To avoid this, align to the Isaac Lab tag that matches the docs you’re using for Isaac Sim 5.1.0:

1. Check out the appropriate Isaac Lab release tag (example name, adjust to the tag NVIDIA indicates for 5.1):

cd /home/leo/Desktop/IsaacLab
git fetch
git checkout v2.3.2  # or the closest 2.3.x tag officially paired with Isaac Sim 5.1.0

2. Re‑install in editable mode:

cd source
pip install -e .

3. Re‑run the import test and the training command.

NVIDIA’s “Getting Started with Isaac Lab” learning materials and the imitation‑learning section describe teleoperation and robomimic‑based workflows at a high level, and they assume this version matching between Isaac Sim and Isaac Lab.⁹⁴³

Dataset path sanity check

The teleoperation and imitation‑learning tutorial you are following demonstrates collecting demonstrations, saving an HDF5 file, then using robomimic to train a policy on a specified task and dataset. As long as:²¹

• generated_dataset_gr1.hdf5 was produced using the same environment configuration (GR1T2) and lives under IsaacLab/datasets, and
• You pass --dataset ./datasets/generated_dataset_gr1.hdf5 from the repo root,

the dataset argument itself is not the cause of the missing module error; that error arises earlier during environment and agent module resolution.¹⁴

---

Putting it all together, to fix your situation on Isaac Sim 5.1.0:

1. Ensure your IsaacLab checkout matches the recommended 2.3.x release for 5.1.0 (e.g., a v2.3.2 tag), rather than an arbitrary main snapshot.⁸⁷³
2. From that checkout, do pip install -e source into the same Python environment you use with Isaac Sim.
3. Confirm isaaclab_tasks.manager_based.manipulation.pick_place.agents imports successfully via ./isaaclab.sh -p -c "import ..." from the repo root.
4. Then re‑run the GR1T2 robomimic training command from the repo root with the same dataset path.

Footnotes

1. https://isaac-sim.github.io/IsaacLab/main/source/overview/imitation-learning/teleop_imitation.html ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

2. https://isaac-sim.github.io/IsaacLab/main/source/overview/imitation-learning/index.html ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

3. https://docs.isaacsim.omniverse.nvidia.com/5.0.0/isaac_lab_tutorials/index.html ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹

4. https://docs.nvidia.com/learning/physical-ai/getting-started-with-isaac-lab/latest/an-introduction-to-robot-learning-and-isaac-lab/03-available-robots-and-environments/02-available-environments.html ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

5. https://isaac-sim.github.io/IsaacLab/ ↩ ↩² ↩³

6. https://docs.isaacsim.omniverse.nvidia.com/5.1.0/installation/requirements.html ↩ ↩²

7. https://forums.developer.nvidia.com/t/isaac-sim-5-1-ga-release/349634 ↩ ↩²

8. https://developer.nvidia.com/blog/streamline-robot-learning-with-whole-body-control-and-enhanced-teleoperation-in-nvidia-isaac-lab-2-3/ ↩ ↩²

9. https://docs.nvidia.com/learning/physical-ai/getting-started-with-isaac-lab/latest/an-introduction-to-robot-learning-and-isaac-lab/01-what-is-robot-learning/04-imitation-learning.html ↩

[LeoE-SJSU]

Ok thank you so much for replying! Sorry I may need a little more hand holding.

Just to test the idea that the isaacsim and isaaclab version could be mismatched, I went ahead and installed a new isaacsim 5.1.0 (through pip and conda) and new isaaclab. 5.3.2
Still no luck.

Using the find source command, I could confirm the pick and place do exist in the folders.

find source -maxdepth 8 -path "*isaaclab_tasks/manager_based/manipulation/pick_place*" -print

I tried running the editable install using pip install -e . in the source folder but I got this error:

ERROR: file:///home/leo/Desktop/IsaacLab/source does not appear to be a Python project: neither 'setup.py' nor 'pyproject.toml' found.

I tried running

./isaaclab.sh -p -c "import isaaclab_tasks.manager_based.manipulation.pick_place.agents as agents; print(agents.__file__)"

But I got the following error:
ModuleNotFoundError: No module named 'pxr'

[LeoE-SJSU]

One more thing:

I can run the training on the Isaac-PickPlace-Locomanipulation-G1-Abs-v0.
But not on the Isaac-PickPlace-GR1T2-Abs-v0 . The only difference between both trainings is the task name and the hdf5 file referenced.

i.e

This one does not run:
./isaaclab.sh -p scripts/imitation_learning/robomimic/train.py \ --task Isaac-PickPlace-GR1T2-Abs-v0 --algo bc \ --normalize_training_actions \ --dataset ./datasets/generated_dataset_gr1.hdf5

This one runs:

./isaaclab.sh -p scripts/imitation_learning/robomimic/train.py \ --task Isaac-PickPlace-Locomanipulation-G1-Abs-v0 --algo bc \ --normalize_training_actions \ --dataset ./datasets/generated_dataset_g1_locomanip.hdf5

[TannerGilbert]

@LeoE-SJSU I'm running into a the problem. Did you find a solution?

@RandomOakForest I ran through your suggested steps and changed to the 2.3.2 and then reinstalled via ./isaaclab.sh --install.

When running ./isaaclab.sh -p -c "import isaaclab_tasks.manager_based.manipulation.pick_place.agents as agents; print(agents.__file__)" I also get: ModuleNotFoundError: No module named 'pxr' . Could this be related to installing via pip instead of binary install (I installed via venv)?

[LeoE-SJSU]

@TannerGilbert I just found the solution.

TLDR:

On the github, rollback to version 2.3.0 and see that the __init__.py file for GR1T2 is different than the current (2.3.2) version.
The file uses a different way to call agents and it seems this way is not being recognized.

To fix the issue roll back the folder to 2.3.0.
If you want to download only the python file (which is what I did), replace the __init__.py found at /IsaacLab/source/isaaclab_tasks/isaaclab_tasks/manager_based/manipulation/pick_place

With the __init__.py found on the Github at

https://github.com/isaac-sim/IsaacLab/blob/release/2.3.0/source/isaaclab_tasks/isaaclab_tasks/manager_based/manipulation/pick_place/__init__.py

---

The agents for these trainings are not python files, they are json files. So agents.py will always return a Module Error.
The troubleshooting steps shown are not useful for this specific issue.

If we look at the train.py error, it happens when it is looking at for an entrypoint, which the name of the agent I believe.
The way to access the agent is found on the __init__.py file.
In 2.3.0, on the G1, and other robots the file reads something like this:

gym.register(
    id="Isaac-PickPlace-GR1T2-Abs-v0",
    entry_point="isaaclab.envs:ManagerBasedRLEnv",
    kwargs={
        "env_cfg_entry_point": pickplace_gr1t2_env_cfg.PickPlaceGR1T2EnvCfg,
        "robomimic_bc_cfg_entry_point": os.path.join(agents.__path__[0], "robomimic/bc_rnn_low_dim.json"),
    },
    disable_env_checker=True,
)

But on 2.3.2 for GR1T2 it reads this:

gym.register(
    id="Isaac-PickPlace-GR1T2-Abs-v0",
    entry_point="isaaclab.envs:ManagerBasedRLEnv",
    kwargs={
        "env_cfg_entry_point": f"{__name__}.pickplace_gr1t2_env_cfg:PickPlaceGR1T2EnvCfg",
        "robomimic_bc_cfg_entry_point": f"{agents.__name__}:robomimic/bc_rnn_low_dim.json",
    },
    disable_env_checker=True,
)

I am not well versed enough into coding to understand why this new version is not working, I only know that they look different and that the old version works.

Anyways, after replacing my __init__.py file I was able to properly train and complete the imitation learning tutorial.

[peterd-NV]

Hi @LeoE-SJSU, @TannerGilbert thank you both for raising this issue and for your detailed notes.

Version 2.3.2 added a change to make the environment registrations to follow a string import style (PR 3803). However, this results in the the pick_place.agents package not being found as it did not have an init.py to denote it as a proper python package. In the prior 2.3.0 version, this did not cause an issue as the training JSON was passed as a direct path.

This problem has been resolved in v3.0.0-beta. I understand that v3.0.0-beta makes a lot of core changes so if you wish to remain on v2.3.2, the error can be fixed by adding an empty file named __init__.py to source/isaaclab_tasks/isaaclab_tasks/manager_based/manipulation/pick_place/agents (here).

You can also reference the __init__.py file here in v3.0.0-beta for an example of the fix.