OpenHUTB/hutb

OpenHUTB / hutb – A cinematic‑grade physics simulator for embodied agents

What it is

An open‑source simulation platform that models people and vehicles (ground, aerial, underwater) with high‑fidelity, movie‑quality physics. It builds on Carla, AirSim, MuJoCo and other engines to give researchers a unified environment for training and testing perception, planning, control, reinforcement‑learning and even large‑model (LLM) pipelines.

Key capabilities

Capability	Details
Embodied entities	Human avatars, ground vehicles, drones, sub‑mersibles, pedestrians, props – all with realistic dynamics.
Multiphysics	Supports Chrono‑based multi‑body simulation, fluid‑dynamics for underwater, and sensor suites (cameras, LiDAR, radar, etc.).
VR cockpit	Optional VR driving cockpit; works with steering‑wheel or keyboard controls.
Air‑ground integration	Switch between Carla road mode and AirSim‑style aerial mode in the same world.
Data synthesis	Built‑in pipelines to generate labeled sensor data for training AI models.
Traffic manager	Advanced traffic control (traffic lights, vehicle spawning, SUMO/PTV Vissim co‑simulation).
Python API	`hutb` wheel packages (Python 3.7‑3.14) expose full simulator control, scene creation, and sensor access.
Extensible plugins	Separate repos for MuJoCo plugin, drone plugin, VR inter‑behavior, etc.
Cross‑platform	Runs on Windows 10+, Ubuntu 18.04+, macOS 12+. Requires a modern CPU, ≥16 GB RAM and an RTX 2070‑class GPU.

How to get started

Download the simulator – run the provided hutb_downloader.exe (or clone the repo).
Install the Python wheel – pip install hutb‑*.whl from hutb/PythonAPI/carla/dist/.
Run an example – e.g. python PythonAPI/examples/generate_traffic.py to populate a town with vehicles and walkers.
Switch modes – use config.py:
- python config.py --map Town10HD?GAME=VR → VR driving cockpit.
- python config.py --map Town10HD?GAME=AIR → drone mode, then python PythonClient/multirotor/hello_drone.py.
Develop – edit source with setup.bat -l (launch editor) and package with setup.bat -p.

Ecosystem & integrations

Carla leaderboard for autonomous‑driving benchmarking.
NVIDIA SimReady / neural rendering / Cosmos for AI‑accelerated rendering.
ROS‑bridge, Scenario Runner, AutoWare, Apollo bridge for connecting real‑world robotics stacks.
Reinforcement‑learning examples and code links.
A library of free digital assets (maps, buildings, vehicles, pedestrians, props) under CC‑BY.

Documentation & community

Full Chinese docs at https://openhutb.github.io (English README also available).
Video tutorials, PPT deep‑dives, and a list of related applications (perception, planning, end‑to‑end, agents, explainability, etc.).

License

Core code: MIT license (non‑commercial clause noted in the badge). Assets: CC‑BY.

Who should use it

Researchers building autonomous‑driving, aerial‑robotics, or underwater‑robotics algorithms.
AI teams needing realistic sensor data for training perception or LLM‑based scene understanding.
Developers who want a single simulator that can handle ground, air and water vehicles together.

All information above is taken directly from the repository’s README; no additional features are inferred.