Risk-informed strategy prescription: An option-level risk framework in the graph model for decision support in human–autonomous vehicle interaction
Year: 2026
Type: article
Abstract: An autonomous vehicle at an unsignalised intersection or a dense merge is exposed to every outcome the interaction can reach. The classical stability definitions of game-theoretic approaches leave this exposure outside the prescription. Accordingly, a Risk-Informed Strategy Prescription (RISP) framework is introduced that converts the ordinal preference orders of the graph model for conflict resolution (GMCR) into an option-level prescription. Every state reachable from the status quo is scored against the decision maker’s own order. The worst score that an option allows when selected, and when rejected, is then compared with a tunable acceptable-risk threshold, which classifies the option as safe, risky, neutral, or frozen. The score is ordinal, every guarantee is stated at that level, and compliance means collision exclusion only below the decision maker’s catastrophic cutoff. The frozen class is the central device: an admissibility condition on each decision maker’s complete strategy gives the frozen-robot failure mode of autonomous decision-making its first computable characterisation within a stability framework. A worst-case threshold-compliance bound is established for an agent that follows the prescription, and a decision-support architecture places the RISP module between opponent modelling and motion planning. The framework is demonstrated on two illustrative three-decision-maker scenarios, an unsignalised intersection and a highway merge. The prescription reduces each interaction to a single option-level recommendation that the classical definitions do not return.
Author: Yasser Talal Matbouli
Institution: King Abdulaziz University
Language: English
Subfield: Human Factors and Ergonomics
Field: Social Sciences
Domain: Social Sciences
Open Access status: gold
Estimated APC paid: $1,990
Funder: King Abdulaziz University
Award: G9001398678
OpenAlex ID: W7236652003
