Research Paper

Core Innovation

This paper introduces KnowVal, which combines a driving knowledge graph encoding traffic laws and ethics with a large language model-based retrieval system and a human-preference-trained value model. This approach surpasses purely data-driven methods by enabling interpretable, value-guided trajectory assessment and improved planning performance.

Why It Matters

Autonomous vehicles face challenges in safely navigating complex traffic scenarios due to limited understanding of driving logic and human values. KnowVal improves safety and reliability by embedding traffic laws and ethical norms into decision-making, enabling scalable adoption across autonomous platforms. This reduces accidents and builds trust in self-driving technology.

Potential Customers & Pain Points

• Autonomous vehicle manufacturers – Need safer more reliable planning
• Fleet operators – Need to reduce accident rates and liability
• Autonomous driving software developers – Need interpretable value-aligned decision models

Business Model

Licensing software modules to autonomous vehicle manufacturers and fleet operators; offering value-aligned planning APIs and consulting for integration.

Competitive Landscape

• Waymo
• Tesla Autopilot
• Cruise
• Aurora

Implementation Challenges

• Integration complexity with existing autonomous systems
• Ensuring real-time performance of knowledge retrieval
• Regulatory approval and validation in diverse driving environments

Validation Strategy

• Pilot deployments with autonomous vehicle OEMs
• Benchmarking on public datasets like nuScenes and Bench2Drive
• User studies to assess value alignment and interpretability
• Regulatory compliance testing and safety certification