FBGA
Personal project, developed during the PhD | 2025 - present
Overview
FBGA - Forward-Backward Generic Acceleration constraints - computes the maximum-performance velocity profile along a path given a complex g-g-v constraint envelope: the combined longitudinal/lateral acceleration limits a vehicle can actually sustain, which change with speed and aren't a simple circle or ellipse. Given a path's curvature profile and that envelope, the forward-backward algorithm finds the fastest velocity profile that never asks the vehicle for more grip than it has, running fast enough for real-time use rather than offline-only optimization.
From 2D to 3D
The library started as a 2D solver - curvature-only path, single g-g envelope - published as Real-Time Velocity Profile Optimization for Time-Optimal Maneuvering With Generic Acceleration Constraints (IEEE Robotics and Automation Letters, 2025). It has since grown a 3D solver, generic over the vehicle's g-g-v model, that accounts for banked and pitched paths, load transfer, and speed-dependent grip - published as Beyond the Apex: Online Minimum-Time Velocity Planning on Three-Dimensional Paths With Variable Friction (IEEE Robotics and Automation Letters, 2026). The 3D solver was originally developed as its own repository, FBGA_3D, and is now ported into FBGA itself as an opt-in module.
Where it runs
Used in the TUM Autonomous Vehicle Systems Laboratory's Indy Autonomous Challenge and Abu Dhabi Autonomous Racing League entries for online velocity planning under dynamic grip constraints.
Status
Public C++ library, maintained solely by me, hosted under the DRIVEWISE organization. 2025 - present.