AutoPreview: A Framework for Autopilot Behavior Understanding
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 introduces AutoPreview, a framework designed to help users build accurate mental models of autonomous vehicle (autopilot) behaviors before deployment. The authors address the problem of expectation mismatches between drivers and autopilot systems, which can lead to distrust and safety risks. Current industry practices, such as reading release notes, are insufficient for calibrating user understanding. AutoPreview solves this by using a "delegate policy" generated via imitation learning to replicate the behavior of a target autopilot. This delegate model outputs explainable, high-level action predictions (e.g., "switch lanes now") rather than executing control commands, allowing users to safely preview how a specific autopilot would behave in real-world scenarios while the human driver retains full vehicle control. The framework is guided by three design principles: safety (preventing dangerous situations during learning), convenience (avoiding burdensome offline introspection), and realism (using real-world interactions). The authors implemented a prototype in the CARLA simulation environment to demonstrate two use cases: previewing new software releases for existing users and comparing autopilot behaviors across different brands for potential buyers. In the prototype, the delegate model translates low-level control actions into high-level explanations, which are presented to the user via an interface that shows what the target autopilot would do if deployed. To evaluate the framework, the authors conducted a pilot study with 10 participants using a between-subject design. Participants were divided into a treatment group (using AutoPreview to infer target behavior via the delegate model) and a comparison group (observing the target autopilot directly controlling the vehicle). The experiment focused on lane-changing behaviors in a simulated two-lane environment. Participants rated the aggressiveness of the driving style and predicted the exact timing of lane changes. Results indicated that the AutoPreview method helped users understand autopilot behavior effectively. The treatment group showed significantly lower timing prediction errors compared to the comparison group, with statistical significance confirmed by Mann-Whitney U and t-tests. Additionally, participants reported high confidence in their predictions and found the framework easy to use. The significance of this work lies in providing a practical tool for improving human-autonomous vehicle interaction. By enabling users to preview and understand autopilot behaviors through a safe, delegate-based interface, AutoPreview helps establish appropriate levels of trust and situational awareness. The findings suggest that indirect previewing via a delegate model is as effective as direct observation for building mental models, offering a scalable solution for consumer education and autopilot evaluation without requiring direct exposure to potentially unpredictable autonomous behaviors.
Key finding
The AutoPreview framework helps users understand autopilot behavior in terms of driving style comprehension, deployment preference, and exact action timing prediction.
Methodology
simulation_modeling
Sample size: 10
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 author_sweep_intake on 2026-05-28.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-28 |
| archive | success | unpaywall | — | — | 2 | 2026-06-04 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-04 |
| chunk | success | chunk | — | — | 1 | 2026-06-04 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-04 |
| enrich | success | — | — | — | 1 | 2026-05-28 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 2 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 15 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
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.
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).
- Methodological Resource: tool software
- Theoretical Contribution: computational model, theory or model