RECTOR: Priority-Aware Rule-Based Reranking for Compliance-Aware Autonomous Driving Trajectory Selection
The paper presents RECTOR, a rule-based reranking system designed for autonomous driving trajectory selection. It prioritizes safety, legal compliance, road conditions, and comfort in its decision-making process. The results demonstrate a significant reduction in safety and legal violations compared to traditional confidence-based selection methods.
- ▪RECTOR uses a tiered rulebook to score trajectory candidates based on safety, legality, road conditions, and comfort.
- ▪The system achieved a reduction in safety and legal violations from 28.58% to 20.42% compared to confidence-only methods.
- ▪Under adversarial conditions, confidence-only selection failed in 100% of scenarios, while rule-aware selectors maintained a 96% success rate.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.25095 (cs) [Submitted on 24 May 2026] Title:RECTOR: Priority-Aware Rule-Based Reranking for Compliance-Aware Autonomous Driving Trajectory Selection Authors:Hadi Hajieghrary, Benedikt Walter, Chaitanya Shinde, Paul Schmitt, Miguel Hurtado View a PDF of the paper titled RECTOR: Priority-Aware Rule-Based Reranking for Compliance-Aware Autonomous Driving Trajectory Selection, by Hadi Hajieghrary and Benedikt Walter and Chaitanya Shinde and Paul Schmitt and Miguel Hurtado View PDF HTML (experimental) Abstract:Autonomous driving stacks must pick one trajectory from a multi-modal candidate set; choosing by model confidence ignores safety, traffic-law, and comfort constraints.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
