Rico Krueger

Description

In the future of urban transportation, human road users (such as pedestrians and cyclists) will likely interact with various autonomous systems (such as self-driving cars and last-mile delivery robots). To operate safely and efficiently, these systems must perceive, reason and act with social intelligence. For example, a self-driving car approaching a pedestrian at the side of the road must be able to predict with high accuracy whether the pedestrian will cross the road (and require the car to slow down to avoid a collision) or yield traffic. The important challenge of how to predict human behaviour in urban environments has been addressed mainly by i) data-driven computer vision approaches and ii) theory-driven discrete choice models. Computer vision approaches excel at exploiting various contextual cues and complex spatiotemporal dependencies. However, these algorithms are often “black-box”, and as such they lack interpretability and transferability, which are crucial for embedding social intelligence into autonomous systems. By contrast, discrete choice models are interpretable and transferrable. Yet, their prediction accuracy is typically low, as their ability to account for contextual information is limited. The lack of powerful behavioural models which simultaneously satisfy the desiderata of high prediction accuracy, interpretability and transferability hampers the advancement of socially aware artificial intelligence for future transportation. The project will address this critical research gap through the development of a new generation of behavioural models at the intersection of econometrics and psychology leveraging machine learning and virtual reality.

Duration

2023-2026

Sponsor

DTU

Lead organisation

DTU

Principal investigator

Rico Krueger

Collaborators (DTU)

• Francisco C. Pereira
• Danya Li

Partners

• Xiao Yu, Aalto University
• Xiang Su, Norwegian University of Science and Technology

[Return to Projects]