In this project we are investigating whether the gaze behavior of a vehicle occupant can be used to detect latent risk perception at an early stage and enable proactive vehicle intervention.

Our central research question is:

Can occupant gaze behavior serve as an early indicator of perceived risk, allowing autonomous vehicles to initiate safety-oriented actions before hazards fully materialize?

This research treats human gaze not merely as a passive observation signal, but as meaningful information for risk assessment.

When an occupant’s gaze becomes fixated on specific objects—such as pedestrians, merging vehicles, or unexpected obstacles—this can be interpreted as an indication of perceived danger.

We propose transforming such gaze cues into a quantitative risk-awareness score that can be directly linked to vehicle-level safety actions, including braking, deceleration, and lane-keeping assistance.

From a learning perspective, we plan to integrate gaze embeddings into end-to-end autonomous driving policies, enabling the vehicle to combine sensor-based perception with human-centered cognitive signals.

This approach shifts the paradigm from purely sensor-driven autonomy to a human-in-the-loop autonomous driving framework where machine perception is augmented by human risk awareness.

• Prof. Dr.-Ing. Borislav Hristov(Projektleitung)

• Chungbuk National University, Dept. of Intelligent Systems & Robotics, South Korea