Improving External Communication of Automated Vehicles Using Bayesian Optimization

TL;DR

This work tackles how to externalize vehicle intent in autonomous cars through eHMIs by applying a HITL multi-objective Bayesian optimization (MOBO) to a VR pedestrian-crossing study with $N=37$ participants. The approach optimizes nine continuous design parameters $p_1$ to $p_9$ across six subjective objectives plus mental demand and time to start crossing, using $q$EHVI with a multi-output GP to obtain Pareto fronts. Findings show MOBO can improve perceived safety and reduce mental demand and crossing time, while trust, predictability, and acceptance improve more modestly, with a robust starting point emerging around cyan visuals, a ~3 Hz flashing rate, and a large front-face eHMI complemented by audible cues. The work yields practical guidelines for multimodal eHMIs, discusses universality versus individual differences, and contributes to standardization efforts for broader adoption and safer pedestrian–AV interactions.

Abstract

The absence of a human operator in automated vehicles (AVs) may require external Human-Machine Interfaces (eHMIs) to facilitate communication with other road users in uncertain scenarios, for example, regarding the right of way. Given the plethora of adjustable parameters, balancing visual and auditory elements is crucial for effective communication with other road users. With N=37 participants, this study employed multi-objective Bayesian optimization to enhance eHMI designs and improve trust, safety perception, and mental demand. By reporting the Pareto front, we identify optimal design trade-offs. This research contributes to the ongoing standardization efforts of eHMIs, supporting broader adoption.

Figures (13)

• Figure 1: The Unity implementation of the crossing scenario from the participant's starting point of view (a) and from above (b).
• Figure 2: The white AV without an eHMI. A cyan light signal near the rearview mirror indicates that it is an AV. In (a), dashed lines show the available width and height for the eHMI mesh generation, including its lower/upper and left/right bounds. (b) Examples of possible mesh positions in allowed regions. The mesh (i.e., its width) is always centered horizontally along the green dashed axis to ensure visibility from both sides. The vertical position can be set along this axis. The eHMI does not cover the front lights (restricted zones; in purple).
• Figure 3: eHMI design parameter value ranges. At the bottom of each column, the default values for the other parameters are shown. Cyan is used as the default color to demonstrate other parameter values.
• Figure 4: Study procedure using HITL MOBO for eHMI design. There were 20 (5 sampling and 15 optimization) iterations.
• Figure 5: Rating for the subjective questionnaires comparing female and male of Pareto-optimal values. The Bayes factor shows trends towards equality (<1) and difference (>1).