Reproduce “Distribution-Free Risk-Aware Planning and Control Under Uncertainty Using Conformal Spectral Risk Control” 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
Junsik Eom, Tulga Ersal
The authors introduce conformal spectral risk control (CSRC) to calibrate a safety margin from prediction errors and use it in MPC. The method extends conformal risk control to spectral risk measures, including CVaR. In 1,000 dynamic-obstacle trials with predictions biased by 0.5 m in both coordinates, violations fell from 52.9% to 6.0% versus SAA-MPC; violations remained and are not collisions.
Key insights
- Spectral risk measures can be reformulated as weighted expectations, allowing their integration into weighted conformal risk control frameworks for distribution-free risk quantification.
- Conservative approximations and offline calibration enable practical, computationally efficient application of conformal methods to control general spectral risk measures in real-time systems.
- Leveraging the Lipschitz continuity of constraint functions, distribution-free prediction sets can be integrated into MPC to provide robust statistical safety guarantees, even when the true uncertainty distribution is unknown or misspecified.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.