Decision-Theoretic Reasoning for Traffic Monitoring and Vehicle Control

Wellman, Michael P.; Russell, Stuart J. · 1995 · ROSA P / United States. Federal Highway Administration

archive: archived pipeline: cataloged verified

Get this paper ↗ (full text — opens at the source; we link to it, we don't host it)

Summary

This report details a joint research project by the University of Michigan and the University of California, Berkeley, funded by the ITS-IDEA program. The study addresses the problem of robust traffic monitoring and automated vehicle control by applying decision-theoretic reasoning and probabilistic models to interpret traffic situations. The primary motivation was to move beyond low-level sensor fusion toward high-level situation assessment, such as recognizing driver intentions and predicting maneuvers, which are critical for intelligent transportation systems, safety analysis, and autonomous driving. The researchers aimed to demonstrate that Bayesian network technology could handle the inherent uncertainty in driver behavior and sensor data in real-time environments. The methodology involved developing and implementing dynamic Bayesian networks (DBNs) using the Hugin probabilistic reasoning system. The team constructed two primary model types: vehicle-centered models to track individual vehicle states (position, velocity) and infer driver intentions, and highway models to capture aggregate traffic variables like flow and congestion. To address computational constraints, the researchers developed "anytime approximation" algorithms, including state-space abstraction and stochastic simulation techniques such as evidence reversal and survival of fittest sampling. These models were tested using stationary highway video footage for visual processing and integrated with the SmartPATH animation system for real-time driving simulations. The experimental design focused on maneuver recognition, where the system inferred high-level plans (e.g., passing, exiting) from partial observations like lane changes and turn signals. The results demonstrated that the proposed probabilistic models were computationally feasible for real-time application on Pentium-class hardware. The vehicle-centered DBNs successfully maintained belief states about vehicle positions and velocities, handling non-Gaussian distributions and incorporating sensor noise. The maneuver recognition framework accurately inferred driver plans; for instance, in a simulation where a vehicle moved to the right lane, the system calculated posterior probabilities for various maneuvers, adjusting its predictions based on contextual evidence such as the presence of blocking vehicles. The anytime approximation algorithms provided increasingly accurate predictions as more computation time was allocated, ensuring robust performance under time constraints. The system successfully fused information from noisy sensors to produce reliable estimates of traffic conditions and individual driver behaviors. The significance of this work lies in confirming the feasibility of using decision-theoretic reasoning for intelligent traffic monitoring and vehicle control. The study concludes that Bayesian networks offer a robust framework for handling uncertainty in traffic interpretation, outperforming traditional stochastic methods in capturing structural uncertainty. The findings support the potential for deploying such technology in applications ranging from near-accident detection and emergency response to autonomous driving. The report highlights that successful deployment depends on reducing sensor costs, but the demonstrated computational efficiency and accuracy suggest that probabilistic reasoning is a viable approach for next-generation intelligent transportation systems.

Key finding

Bayesian network models successfully inferred driver plans and predicted future vehicle positions in real-time using partial observations and anytime approximation algorithms.

Methodology

simulator

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via bulk_ingest_rosap on 2026-05-23 (6 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• situational awareness
• mental model of traffic
• anticipation
• gap acceptance

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Theoretical Contribution: computational model, theory or model