Enhancing Autonomous Driving Safety with Collision Scenario Integration

2025-03-05

AI Summary

Collision data is scarce in autonomous driving planning, and existing imitation learning methods struggle to effectively model hazardous scenarios. Method: This paper proposes SafeFusion—a novel training framework—and CollisionGen, a generative data pipeline. CollisionGen enables large-scale synthesis of high-fidelity collision scenarios guided by natural language prompts, refined via rule-based filtering and safety-aware joint optimization. SafeFusion integrates generative modeling, language-grounded scene descriptions, safety-oriented loss functions, and end-to-end planning training. Contribution/Results: The approach improves planning success rate by 56% in high-risk scenarios while preserving baseline performance on routine driving tasks. It significantly enhances system robustness and cross-scenario generalization capability, marking the first work to achieve prompt-driven, rule-constrained, and safety-optimized generation of collision-rich training data for autonomous driving planners.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Natural Language Processing: Safety and Robustness
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

Autonomous vehicle safety is crucial for the successful deployment of self-driving cars. However, most existing planning methods rely heavily on imitation learning, which limits their ability to leverage collision data effectively. Moreover, collecting collision or near-collision data is inherently challenging, as it involves risks and raises ethical and practical concerns. In this paper, we propose SafeFusion, a training framework to learn from collision data. Instead of over-relying on imitation learning, SafeFusion integrates safety-oriented metrics during training to enable collision avoidance learning. In addition, to address the scarcity of collision data, we propose CollisionGen, a scalable data generation pipeline to generate diverse, high-quality scenarios using natural language prompts, generative models, and rule-based filtering. Experimental results show that our approach improves planning performance in collision-prone scenarios by 56% over previous state-of-the-art planners while maintaining effectiveness in regular driving situations. Our work provides a scalable and effective solution for advancing the safety of autonomous driving systems.

Problem

• Improves autonomous vehicle safety using collision data integration.
• Addresses scarcity of collision data with scalable generation pipeline.
• Enhances planning performance in collision-prone scenarios by 56%.

Innovation

• SafeFusion integrates safety metrics for collision avoidance
• CollisionGen generates diverse collision scenarios using AI
• Improves planning in collision-prone scenarios by 56%