Third-party cyber evaluations involving OpenAI models
openai.com·14h ago
AI Research
arXiv CS.AI·May 26·5 min read
RECTOR: Priority-Aware Rule-Based Reranking for Compliance-Aware Autonomous Driving Trajectory Selection
TL;DR
Autonomous vehicles often struggle to select safe and compliant driving trajectories from multiple options. RECTOR (Rule-Enforced Constrained Trajectory Orchestrator) is a reranking tool that evaluates trajectories based on a prioritized rule set focusing on safety, legality, road conditions, and comfort.
✦ Why It Matters
Engineers can implement RECTOR to improve safety and compliance in autonomous driving applications without extensive model retraining.
Key Takeaways
- Adopt RECTOR for improved trajectory selection in autonomous vehicles.
- Utilize a tiered rulebook to prioritize safety and legality.
- Reduce safety violations significantly compared to confidence-based methods.
- Ensure compliance without retraining existing prediction models.
How It Works
RECTOR employs a post-generation reranking mechanism that scores candidate trajectories based on a structured rulebook. This rulebook prioritizes safety over legal considerations, which in turn is prioritized over road conditions and comfort.
By using differentiable proxies, RECTOR can evaluate the applicability of these rules in real-time scenarios, allowing for a deterministic selection process that respects the established priorities.