Reproduce “Kinodynamic Trajectory Following with STELA: Simultaneous Trajectory Estimation & Local Adaptation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Edgar Granados, Sumanth Tangirala, Kostas E. Bekris
STELA is a graphical optimization framework that unifies trajectory estimation and control for kinodynamic systems, robustly handling sensing noise, actuation errors, and model discrepancies. It achieved high success rates (often 100%, consistently >80-90% at high noise) in simulations and real-world experiments on a non-linear robotic car, demonstrating adaptation to unmodeled features like ramps and operating at 20-30Hz.
Key insights
- Treating the duration of execution (∆t) between states as an optimization variable within the factor graph allows dynamic adaptation (stretching/contracting segments) to compensate for noise and model errors.
- Initializing the factor graph with a high-quality, feasible trajectory from an asymptotically optimal sampling-based motion planner (SBMP) is crucial for overcoming local minima and ensuring robust performance in complex environments.
- A novel sliding window mechanism, comprising both a past history for smoothing and a forward horizon for adaptation, combined with incremental inference (iSAM2), enables efficient, high-frequency (20-30Hz) online updates.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.