Toward Informed AV Decision-Making: Computational Model of Well-being and Trust in Mobility

TL;DR

The paper presents a Dynamic Bayesian Network to infer user well-being $w_k$, trust $t_k$, and intention $i_k$, as well as others’ well-being $w^O_k$, from dyadic AV–user interactions and to integrate these states into AV control via a causal inference model (CIM). An observational Wizard-of-Oz study with 300 participants informs structure learning and parameter estimation, yielding DBN models (SC-DBN and RC-DBN) that predict latent states with notable accuracy ($77\%$ for well-being, $67\%$ for trust, $95\%$ for intention) and reveal dynamics under action alignment. The CIM-based decision policy optimizes AV actions to balance user well-being, trust, others’ well-being, and action costs, with VOI analyses indicating which past variables most improve decision quality. The work advances human-centered automation by enabling real-time, uncertainty-aware reasoning about cognitive states to guide safer and more acceptable AV behavior; future work should address longitudinal data and broader multi-user environments.

Abstract

For future human-autonomous vehicle (AV) interactions to be effective and smooth, human-aware systems that analyze and align human needs with automation decisions are essential. Achieving this requires systems that account for human cognitive states. We present a novel computational model in the form of a Dynamic Bayesian Network (DBN) that infers the cognitive states of both AV users and other road users, integrating this information into the AV's decision-making process. Specifically, our model captures the well-being of both an AV user and an interacting road user as cognitive states alongside trust. Our DBN models infer beliefs over the AV user's evolving well-being, trust, and intention states, as well as the possible well-being of other road users, based on observed interaction experiences. Using data collected from an interaction study, we refine the model parameters and empirically assess its performance. Finally, we extend our model into a causal inference model (CIM) framework for AV decision-making, enabling the AV to enhance user well-being and trust while balancing these factors with its own operational costs and the well-being of interacting road users. Our evaluation demonstrates the model's effectiveness in accurately predicting user's states and guiding informed, human-centered AV decisions.

Figures (7)

• Figure 1: R-DBN (blue) and O-DBN (red) in general form in alternative order.
• Figure 2: Four scenarios of ego-delivery robot interaction. In scenarios $S1$, $S2$, and $S3$, the roles of contributor and receiver are interchanged depending on who takes the yielding or unyielding action first. In scenario $S4$, the scooter acts as the contributor while the robot is the receiver. The robot and scooter switch locations in the robot-contributor scenario of $S4$.
• Figure 3: Web-based riding environment.
• Figure 4: Final model structure of R-DBN (blue) and O-DBN (red)
• Figure 5: Expected values of inferred states over $10$ events using the model given the scooter's actions and action alignment.