Efficiently Explained: Leveraging the SEEV Cognitive Model for Optimal Explanation Delivery

Bairy, Akhila; Fränzle, Martin · 2024 · Crossref

DOI: 10.54941/ahfe1005221

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 optimizing the timing and frequency of explanations provided by autonomous vehicles (AVs) to human occupants. While existing research has extensively covered the content of explanations, there is a significant gap in understanding the precise temporal dynamics of explanation delivery. The authors argue that both providing and skipping explanations induce cognitive workload: presenting explanations requires processing effort, while omitting them may trigger active environmental scanning by passengers to understand unexpected vehicle behavior. The goal is to minimize this total cognitive workload by determining the optimal moment to deliver explanations, balancing the need for information against the risk of redundancy or memory overload. To solve this, the authors develop a probabilistic reactive game model, specifically a Markov Decision Process (MDP), integrating the SEEV (Salience, Effort, Expectancy, Value) cognitive attention model. The SEEV model predicts the probability of a human paying attention to an information item based on bottom-up factors (salience, effort) and top-down factors (expectancy, value). In this framework, the AV’s explanation mechanism acts as a strategic player aiming to minimize expected workload, while the human’s attention dynamics are modeled as a random player governed by SEEV probabilities. The model was implemented in MATLAB using backward Bellman induction to calculate the minimum workload for various scenario durations. The study focuses on a specific scenario where an AV stops at a green light to yield to an emergency vehicle, analyzing the optimal timing for explaining this action. The experimental results demonstrate that optimal explanation timing is not simply "as early as possible" but follows a piecewise affine function relative to the scenario duration. For short horizons (2–5 seconds), the optimal time to provide an explanation is 2 seconds before the event. As the horizon extends to 6–10 seconds, the optimal timing shifts later, ranging from 3 to 7 seconds before the event. For durations up to 15 seconds, the optimal time stabilizes at 3 seconds prior to the event. However, for horizons of 16 seconds or longer, the model suggests providing two explanations: one at the start and another 3 seconds before the event, to account for working memory limitations. The computational analysis indicates that for scenarios up to 20 seconds, the optimization can be performed online in under one second, though computation time increases exponentially for longer horizons. The significance of this work lies in providing a formal, algorithmic approach to explanation timing that reduces cognitive load in human-machine interaction. By leveraging game theory and cognitive modeling, the authors offer a method to dynamically adjust explanation delivery based on real-time context. The authors note that current cost values are heuristic and plan to validate them with empirical psychological data. Future work will extend the model to multi-passenger scenarios and integrate semantic content decisions, contributing to safer and more trustworthy AV systems.

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.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

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

• situational awareness

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: theory or model, computational model