BROWSE
Detail
Conference Paper
Predicting Passenger Anxiety with a Spatio-Temporal PIGNN in VRAutonomous Driving
Cited 0 time in 
Share 
Share
- Authors
- Mi Chang, Jiwoo Han, Woojin Kim, Daesub Yoon
- Issue Date
- 2025-12
- Citation
- ACM SIGGRAPH Asia (SA) 2025, pp.1-2
- Language
- English
- Type
- Conference Paper
- Abstract
- Understanding passenger emotions across different autonomous driving modes is essential for designing adaptive in-vehicle systems. We present a spatio-temporal physics-informed graph neural network (PIGNN) to classify passenger anxiety in a virtual reality (VR) autonomous driving environment. Our model incorporates five temporal constraints (e.g., gaze velocity) and four spatial constraints (e.g., lateral motion), each processed through separate graph neural network (GNN) branches. Among the three model configurations, the attention-weighted multi-constraint architecture achieved the highest accuracy (96.6%) and the lowest test loss (0.5) in classifying passenger anxiety across baseline and tense driving scenarios. By integrating domain-specific physics-informed constraints, our model captures temporally and spatially evolving emotional patterns within physically plausible boundaries, enabling more realistic classification of passenger anxiety.
- KSP Keywords
- Domain-specific, Emotional patterns, Multi-constraint, Virtual Reality, autonomous driving, driving environment, driving mode, in-vehicle system, lateral motion, neural network(NN), spatial constraints