Researchers model human emotions in AI driving models
New AI driving model predicts human emotions in real-time using valence and arousal
A team of researchers from TU Delft, Volvo Cars, and other institutions published a paper introducing the first active inference model of human driving that incorporates emotional states. The work, titled 'Emotion in an active inference model of human driving' and submitted to arXiv in June 2026, extends prior research by modeling emotions along the circumplex model dimensions of valence and arousal within a continuous-state driving environment.
The researchers conditioned affective estimates on both current states and predicted future outcomes, evaluating the approach in two interactive driving scenarios. Their results showed that the generated emotion signals corresponded to affective patterns reported in similar real-world scenarios. This represents a significant step toward more human-like AI systems that can anticipate not just actions but emotional responses in complex environments like autonomous vehicles.
The study bridges active inference—a framework for modeling adaptive behavior—and affective computing, with potential applications in autonomous driving, human-robot interaction, and AI safety.
- Researchers from TU Delft and Volvo Cars built the first active inference model of driving that incorporates human emotions using valence and arousal metrics
- The model predicts emotional responses in continuous driving scenarios by considering both current states and future outcomes in two interactive test cases
- Results showed alignment between model-generated emotions and real-world affective patterns in driving situations
Why It Matters
Could lead to more human-like AI systems that better predict and respond to human emotional states in safety-critical applications like autonomous vehicles