Newton-Raphson Flow for PX4
Links:
ACC 2024 Paper (PDF) TCST 2025 Paper (PDF) Hardware Experiments Video View Python Version View C++ Version
Hardware Experiments
Full hardware demonstrations from the TCST 2025 paper, flown in Georgia Tech’s Indoor Flight Lab on a Holybro X500V2 with a Raspberry Pi 4 doing all onboard computation. Includes NR Flow helix, yawing-circle, and yawing-helix runs on the quadrotor, plus the miniature blimp, alongside the NMPC baseline for comparison.
NR Flow reframes trajectory tracking as an optimization problem solved at each timestep by the Newton-Raphson method — yielding fast, accurate control that is more computationally efficient than NMPC while matching or exceeding its tracking performance. The result: a controller that runs comfortably on a Raspberry Pi 4 onboard a real quadrotor.
Integral Control Barrier Functions (I-CBFs) are baked in to handle actuation limits without discontinuous switching, and all computations are JAX JIT-compiled for real-time deployment.
Key features:
- Newton-Raphson iterative Jacobian inversion for feedback linearization
- Integral CBF safety constraints on actuator inputs (enabled by default, configurable)
- JAX JIT-compilation for real-time performance on resource-constrained hardware
- Structured CLI for simulation and hardware deployment across multiple trajectory types
- Optional CSV logging with automatic analysis notebook generation
- Demonstrated on a Holybro X500V2 + Raspberry Pi 4
Academic foundation:
| Venue | Year |
|---|---|
| American Control Conference (ACC) | 2024 |
| IEEE Transactions on Control Systems Technology (TCST) | 2025 |
Built with: Python · JAX · ROS 2 · PX4