What is it about?
At a road intersection, the trajectories of multiple vehicles cross each other. Where those trajectories overlap at the same moment, there is a risk of collision, although this can prevented by adjusting the speed or curvatures, partly by social interaction. Yet, this adjustment also effects other vehicles on the same crossing, changing the situation. Typical trajectories can be learned from recorded dataset, but in most cases low-risk trajectories dominates the dataset, while most can be learned from high-risk situations. With a self-learning attention mechanism the predictions for high-risk sitations can be improved.
Featured Image
Photo by Peter Hofland on Unsplash
Why is it important?
This work describes a combination of planning and prediction, which is needed to be navigate complex road intersections often seen in urban environments.
Perspectives
This work uses several state-of-the-art deep-learning techniques, like transformers, long short-term memory & gated recurrent unit networks, but at the end its the self-learned attention model that gives the final boost.
Dr. Arnoud Visser
Universiteit van Amsterdam
Read the Original
This page is a summary of: Uncertainty-aware risk assessment and GRU-based risk level map generation for RSU-assisted vehicles, Accident Analysis & Prevention, October 2026, Elsevier,
DOI: 10.1016/j.aap.2026.108649.
You can read the full text:
Contributors
The following have contributed to this page
