MPPI with Control Barrier Functions for F1/10: Robust Safety Under Real-World Uncertainty
May 7, 2025
·
1 min read
This project implements Shield-MPPI, a novel integration of Control Barrier Functions (CBFs) with Model Predictive Path Integral (MPPI) control, on the F1/10 autonomous racing platform to achieve robust, safe navigation under real-world uncertainty.
Key contributions include:
- Safety Assurance: Enforced via discrete-time CBFs ensuring forward invariance of a predefined safe set.
- Cost Augmentation and Control Filtering: Augmented MPPI trajectory costs with CBF penalties and applied gradient-based filtering to guarantee real-time safety.
- Robustness Evaluation: Assessed system under disturbances, noise, and model mismatch in simulated and physical racing environments.
- Computational Feasibility: Validated Shield-MPPI’s real-time performance on resource-limited platforms.
The approach demonstrates significant improvement over baseline MPPI in terms of collision avoidance, track adherence, and robustness.
