Summary
This paper addresses a critical challenge for autonomous vehicles navigating occluded intersections — where buildings or other vehicles block the view of potential hazards like phantom vehicles suddenly entering the intersection.
The Challenge: Traditional planning methods either ignore occluded risks (leading to collisions) or adopt overly conservative behaviors (severely compromising travel efficiency). Real-time replanning in such environments is computationally intensive due to non-convex constraints.
Our Approach: We developed an occlusion-aware contingency planning framework that:
- Uses reachability analysis to quantify risks from occluded phantom vehicles, generating dynamic velocity boundaries
- Simultaneously optimizes two trajectories: an exploration trajectory for situational awareness, and a safety fallback trajectory as a safe alternative
- Shares a common initial segment between both trajectories to ensure smooth transitions
- Employs consensus ADMM to decompose the complex biconvex optimization into low-dimensional convex subproblems solved in parallel
Key Results:
- 30.2% reduction in intersection traversal time compared to baseline methods
- Maintains minimum velocity of 1.64 m/s (versus near-zero for baselines)
- 44.8% faster computation than ST-RHC, 90.9% faster than Control-Tree
- Real-time performance (23.8 ms average solve time) validated on 1:10 scale robotic platform
The framework enables autonomous vehicles to navigate occluded intersections safely while maintaining travel efficiency, scaling effectively with increasing numbers of surrounding vehicles.