The AUSM Lab has started a new SmartTO project with the industry partner AlphaEarth Connect (AEC). The project began on March 15, 2026. It is led by Prof. Gunho Sohn and Prof. Manos Papagelis (Electrical Engineering and Computer Science, York University).
The goal is to predict crash risk for each segment of a road network. Some roads are much more dangerous than others. If we can find them early, we can make them safer.
The team uses a graph neural network (GNN). A GNN treats the road network as a graph: each road segment is a node, and segments that meet at an intersection are linked. This lets the model learn from a road's neighbours, not just the road itself.
Each road segment is described by three kinds of features:
- AlphaEarth satellite embeddings: compact summaries of what the land around the road looks like from space.
- Terrain features: slope, curvature and elevation from digital elevation models.
- Road attributes: road class, number of lanes, speed limit and segment length.
The project first reproduces a published GraphSAGE road-safety model (Gao et al., 2025). It then adapts the model to AEC's large crash and road datasets, using York's high-performance computing resources. The team will also test whether the model works across the whole continental U.S. highway network.
Ph.D. student Mohammad Moein Sheikholeslami leads the model work. Undergraduate student Arpan Sandhu builds the data pipeline. AEC engineers give weekly technical mentoring. The results will be prepared for a joint publication.
SmartTO is a York University hub, run with Centennial College, that helps startups and small businesses develop and test smart mobility technologies. Learn more about SmartTO.