Multipolicy Decision-Making for Autonomous Driving via Changepoint-based Behavior Prediction
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This paper addresses the challenge of decision-making for autonomous vehicles in dynamic traffic environments, specifically focusing on the uncertainty regarding the intentions of other traffic participants. Existing methods often rely on hand-tuned heuristics or numerical optimization that fail to capture coupled dynamic interactions, or they employ Partially Observable Markov Decision Process (POMDP) solvers that are computationally intractable for real-world scenarios. The authors propose an integrated approach that combines behavioral inference with decision-making by modeling vehicle behavior as a discrete set of closed-loop policies. This allows the system to anticipate the uncertain intentions of nearby cars and evaluate the consequences of potential actions through coupled interactions. The methodology consists of two interleaved stages: behavioral prediction and decision-making. For prediction, the system employs Bayesian changepoint detection (specifically the CHAMP algorithm) on the observed history of states for nearby vehicles. This technique segments the history to infer the probability distribution over potential policies (e.g., lane following, lane changing, turning) that each vehicle might be executing. It also includes an anomaly detection mechanism based on policy likelihood and segmentation ambiguity to identify erratic behavior. For decision-making, the system samples from these inferred policy distributions to generate high-likelihood scenarios of coupled interactions. It then performs closed-loop forward simulations of these samples to evaluate outcomes using a user-defined reward function that considers safety, comfort, and goal accomplishment. The autonomous vehicle executes the policy that maximizes the expected reward. The authors evaluate the system using real-world traffic-tracking data collected from a 2013 Ford Fusion autonomous vehicle platform equipped with LIDAR and inertial navigation sensors, as well as simulated highway traffic scenarios. Results demonstrate that the multipolicy sampling strategy generates high-likelihood samples of coupled vehicle interactions significantly faster than uninformed sampling strategies. The changepoint-based approach effectively infers the likely intentions of other agents and detects anomalous behavior. In simulation experiments, such as a four-way stop intersection, the system successfully evaluated outcomes where some vehicles proceeded while others yielded, validating the tractability and effectiveness of the approach. The significance of this work lies in providing a computationally efficient framework for autonomous driving decision-making that accounts for the closed-loop interactions between agents. By leveraging domain knowledge to define a finite set of policies and using changepoint detection to anticipate behavior, the system avoids the scalability issues of general POMDP solvers and the limitations of heuristic methods. This enables robust decision-making in complex, uncertain traffic environments, allowing the autonomous vehicle to react appropriately to the predicted intentions and actions of other participants.
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.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | OpenAlex-citations | — | — | 1 | 2026-06-25 |
| archive | success | unpaywall | — | — | 2 | 2026-06-26 |
| extract | success | cached | — | — | 2 | 2026-06-26 |
| clean | success | clean | — | — | 1 | 2026-06-25 |
| chunk | success | chunk | — | — | 1 | 2026-06-25 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-25 |
| promote | success | — | — | — | 1 | 2026-06-25 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 1 | 2026-06-26 |
| tag | success | vector_similarity | — | — | 6 | 2026-06-25 |
| verify | success | — | — | — | 1 | 2026-06-26 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
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