Risk-Based Filtering of Valuable Driving Situations in the Waymo Open Motion Dataset

2025-06-29

AI Summary

Efficiently identifying high-value interactive driving scenarios from massive autonomous driving datasets remains challenging for safety-critical testing. Method: This paper proposes a risk-aware scenario filtering approach that introduces a novel risk propagation model integrating both first-order (direct) and second-order (indirect) vehicle interactions—thereby overcoming the limitations of conventional methods relying solely on immediate interaction metrics. Leveraging probabilistic risk inference, the method jointly evaluates primary (directly interacting) and secondary (indirectly influencing) vehicle relationships to score and prioritize scenarios. Results: Experiments on the Waymo Open Motion Dataset demonstrate that the method effectively identifies diverse, complementary high-risk interactive scenarios. The curated benchmark dataset is publicly released, significantly enhancing test scenario data quality and coverage effectiveness for autonomous driving validation.

Technology Category

• Reasoning under Uncertainty: Relational Probabilistic Models
• Search and Optimization: Sampling/Simulation-based Search
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Improving automated vehicle software requires driving data rich in valuable road user interactions. In this paper, we propose a risk-based filtering approach that helps identify such valuable driving situations from large datasets. Specifically, we use a probabilistic risk model to detect high-risk situations. Our method stands out by considering a) first-order situations (where one vehicle directly influences another and induces risk) and b) second-order situations (where influence propagates through an intermediary vehicle). In experiments, we show that our approach effectively selects valuable driving situations in the Waymo Open Motion Dataset. Compared to the two baseline interaction metrics of Kalman difficulty and Tracks-To-Predict (TTP), our filtering approach identifies complex and complementary situations, enriching the quality in automated vehicle testing. The risk data is made open-source: https://github.com/HRI-EU/RiskBasedFiltering.

Problem

• Identify valuable driving situations in large datasets
• Detect high-risk situations using probabilistic risk model
• Enrich automated vehicle testing with complex scenarios

Innovation

• Risk-based filtering for valuable driving situations
• Probabilistic risk model detects high-risk scenarios
• Considers first and second-order risk situations