Skip to content
breuerpeterPublic

Latest commit

 

History

153 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

The nexus logo: flight stack simulation with Newton



PyPI Docs

lint cpu-pytest gpu-pytest docs codecov

Newton Isaac Sim Isaac Lab Rerun Python

Built on Newton, Warp and OpenUSD, nexus lets you develop drone subsystems and test them as part of the full flight stack.

What nexus adds to Newton

  • One deterministic loop. Each tick runs the sensors, the flight stack and Newton's step, always in that order. On a GPU the tick replays as CUDA graphs. On CPU a flight with the Proportional Integral Derivative (PID) controller repeats bit for bit, which a CI gate checks. Execution
  • Components you can swap. Sensors, estimators, guidance, controllers, and force elements fill fixed roles under one contract. Write your own, declare it in the vehicle's Universal Scene Description (USD) file beside Newton's physics, and the loop runs it as it runs the shipped ones. Your own component
  • PX4 in lockstep. PX4 Software In The Loop (SITL) flies the vehicle in MAVLink lockstep with the loop. QGroundControl and offboard scripts connect through the run's links. Running SITL flights
  • RTX sensors from Isaac Sim. A vehicle that declares a camera or lidar starts an Isaac Sim container beside the loop. It renders each frame at the loop's poses, and the loop never runs inside Isaac Sim. RTX cameras and lidar
  • One vehicle file, three uses. PX4 flies the vehicle's USD file in CI, Isaac Lab trains a policy on the same file, and the design examples take gradients through Newton from it. A catalog pins each vehicle and scene by its hash. Examples
  • Every tick recorded. The Recorder keeps every row of a run, and the Logger writes it to a Rerun recording. Logging

Installation

The framework requires Python 3.12, Linux, and a CUDA-capable GPU. It ships on PyPI. Install it with uv, which fetches Python 3.12 if the machine lacks it. If you don't have uv, install it first:

uv venv --python 3.12
uv pip install nexus-uav

Optional extras: policy adds PyTorch, examples adds jerk-limited trajectories, acados adds Nonlinear Model Predictive Control (NMPC) code generation, and all equals policy plus examples.

For development, clone the repo and sync it:

git clone https://github.com/breuerpeter/nexus.git
cd nexus
uv sync                 # install the framework
uv run nexus run --vehicle astro_max_base --scene empty   # fly a vehicle against PX4 SITL

Quickstart

Run the command-line tool with a vehicle and a scene. astro_max_base declares PX4 as its controller, and empty is flat ground:

uv run nexus run --vehicle astro_max_base --scene empty

Run a self-contained example. Every example records a Rerun .rrd, and --list names them:

uv run -m nexus_uav.examples sampling_mpc

The decoupled SITL flight workflow, nexus plus PX4 SITL plus QGroundControl, with the port map, lives in Running a SITL flight.

Documentation

The full docs, with the design, the guide, the examples and the API reference, are at https://breuerpeter.github.io/nexus/.

Contributing

Contributions are welcome. See Contributing for setup and the conventional-commit / pre-commit workflow, and Conventions for the framework-versus-example rule.

License

Licensed under the Apache License 2.0. © 2026 Peter Breuer

The hosted vehicle and scene assets and the Astro photograph in the banner aren't Apache-licensed. See License.