habitat-sim Wheels for TBP Monty

Experimental pip/UV-compatible wheels for habitat-sim, built specifically for the Thousand Brains Project (TBP) Monty sensorimotor learning system.

⚠️ Unofficial community build - Not affiliated with Meta/FAIR. For production use, see official habitat-sim.

What is This?

TBP Monty is a sensorimotor learning system that learns through active sensing (like your brain!). It uses habitat-sim for realistic 3D environments in object recognition experiments.

Problem: Official habitat-sim requires conda, incompatible with modern UV/pip workflows.

Solution: Standard Python wheels on PyPI - install with pip or uv, no conda needed.

Installation

# With pip
pip install habitat-sim-uv-wheels-experimental

# With uv (recommended - much faster!)
uv pip install habitat-sim-uv-wheels-experimental

# Import as usual
python -c "import habitat_sim; print(habitat_sim.__version__)"

For TBP Monty Users

# Clone TBP Monty
git clone https://github.com/thousandbrainsproject/tbp.monty.git
cd tbp.monty

# Install everything with UV (includes habitat-sim)
uv sync --extra dev

# Run an experiment
uv run monty -e surf_agent_1lm_10distobj_base

See full TBP installation guide.

WSL2 GPU Support

For hardware-accelerated rendering on WSL2:

# Enable D3D12 backend (NVIDIA RTX recommended)
export HSIM_DISABLE_CUDA_DEVICE=1
export GALLIUM_DRIVER=d3d12
sudo chmod 666 /dev/dri/renderD128

# Run with GPU
HSIM_DISABLE_CUDA_DEVICE=1 GALLIUM_DRIVER=d3d12 python your_script.py

Platform Support

• Linux: ✅ manylinux2014+ (Ubuntu 20.04+, Debian 11+, etc.)
• WSL2: ✅ With D3D12 GPU support
• macOS: ❌ Not yet available
• Windows: ❌ Use WSL2

Python Versions:

• Python 3.8: ✅ Available
• Python 3.9+: 🚧 Coming soon (request here)

What's Different from Official?

This build:

• ✅ Available on PyPI (no conda required)
• ✅ WSL2 GPU support via HSIM_DISABLE_CUDA_DEVICE=1
• ✅ Tested with TBP Monty experiments
• ✅ Same code as official v0.2.2 (only packaging differs)

Quick Example

import habitat_sim

# Load a 3D scene
backend_cfg = habitat_sim.SimulatorConfiguration()
backend_cfg.scene_id = "path/to/scene.glb"

# Create agent configuration
agent_cfg = habitat_sim.agent.AgentConfiguration()

# Initialize simulator
cfg = habitat_sim.Configuration(backend_cfg, [agent_cfg])
sim = habitat_sim.Simulator(cfg)

# Get an observation
obs = sim.get_sensor_observations()
rgb = obs["color_sensor"]
print(f"Image shape: {rgb.shape}")

For TBP Monty examples, see TBP documentation.

Troubleshooting

Import Error after Install

Problem: pip install habitat-sim-uv-wheels-experimental succeeded but import habitat_sim fails.

Solution: The PyPI package name is different from the import name (this is normal). Just import habitat_sim.

Rendering Errors on WSL2

Problem: Crashes or black screens when rendering.

Solution: Set WSL2 environment variables (see "WSL2 GPU Support" above).

Missing Magnum Module

Problem: ModuleNotFoundError: No module named 'magnum'

Solution: Magnum should be installed automatically. Try reinstalling:

pip install --force-reinstall habitat-sim-uv-wheels-experimental

Links

TBP Monty:

• Documentation: https://thousandbrainsproject.readme.io/
• Repository: https://github.com/thousandbrainsproject/tbp.monty
• Community: https://thousandbrains.discourse.group/

This Fork:

• Repository: https://github.com/killerapp/habitat-sim
• Build Guide: UV_WSL2_BUILD.md
• TBP Instructions: README_TBP_USERS.md

Original habitat-sim:

• Repository: https://github.com/facebookresearch/habitat-sim
• Official Docs: https://aihabitat.org/docs/habitat-sim/

Contributing

• TBP Monty issues: https://github.com/thousandbrainsproject/tbp.monty/issues
• Wheel build issues: https://github.com/killerapp/habitat-sim/issues
• Official habitat-sim: https://github.com/facebookresearch/habitat-sim/issues

License

MIT License - Same as official habitat-sim.

Credits: All credit for habitat-sim goes to the original FAIR team at Meta AI. This is just alternative packaging.

---

Built for the Thousand Brains Project community 🧠