Simulation study demonstrates adaptive driving behavior in autonomous vehicles, highlighting foundation model integration with passenger cognitive states.
Key Points
To develop a human-centered autonomous driving architecture that personalizes vehicle dynamics and resolves ambiguous navigation scenarios by interpreting both external traffic and internal occupant states.
Formulated PACE-ADS using three interacting foundation model agents: a Driver Agent for external road perception, a Psychologist Agent for occupant state inference (facial expressions and verbal instructions), and a Coordinator Agent for semantic planning.
Evaluated system behavior in closed-loop CARLA driving simulations under prescribed occupant-state trajectories, verbal commands, and operational vehicle immobilization conditions.
PACE-ADS selectively adapted driving kinematic dimensions associated with ride comfort in response to detected changes in passenger psychological state and cognitive instructions.
The system successfully resolved and recovered from vehicle immobilization scenarios using foundation model self-reasoning combined with occupant-in-the-loop guidance.