ODConNet: OD Pairs based Conditional Travel Time Inference Network

ICLR 2026 Conference Submission14476 Authors

Keywords: Navigation, Travel Time, Origin-Destination Pair, Cell Segmentation, Conditional Network

Abstract: Estimating travel time in navigation systems enhances user decisions. Current path-based models face high computational costs due to their reliance on path-specific calculation. In this paper, we propose ODConNet(OD Pairs based Conditional Travel Time Inference Network), a model that leverages origin-destination pairs to reduce estimation time. ODConNet employs multi-scale cells to model traffic networks, learning common routes and travel times simultaneously. As a result, the model trained on M(suburb-scale) and S(intersection-scale) cells achieved an inference time of 0.2ms per trajectory, with MAE of 4.3minutes and MAPE of 20.9%. This approach enhances the memory efficiency of navigation systems and improves traffic-based search service.

Primary Area: generative models

Submission Number: 14476