Conformal Prediction for Risk-Aware Robot Navigation
A new method for risk-aware navigation in unknown environments uses conformal prediction to handle sensor noise and generate obstacle ellipsoids. The approach, presented in a paper on arXiv, introduces two nested differentiable optimization layers to construct control barrier functions for obstacle avoidance and feasibility. A quadratic program-based safety-critical control law integrates these constraints. Numerical simulations demonstrate the framework's effectiveness.
Key facts
- Method uses conformal prediction for risk-aware obstacle ellipsoids
- Two nested differentiable optimization layers build control barrier functions
- Quadratic program-based safety-critical control law integrates constraints
- Numerical simulations validate the approach
- Paper submitted to arXiv on May 28, 2025
- Addresses uncertainty from sensor noise
- Focuses on autonomous vehicles in urban systems
- Elliptical-shaped robot model used
Entities
Institutions
- arXiv