Ego-Vehicle Speed Prediction with Walk-Ahead

Intelligent vehicle technology is becoming more and more important to counteract high emission levels and climate change while at the same time maintaining a reliable and accessible transportation system. To this end, ego-vehicle speed prediction, i.e., the forecasting of the own vehicle speed in the near future, has emerged as an important research direction for enabling advanced driver assistance systems. In this paper, we propose the WHEELS (Walk-Ahead Ego Vehicle Speed) model for ego-vehicle speed prediction. Our WHEELS combines vehicle speed information retrieved with contextual information such as road network information, weather information, or the day of the week. At the core, WHEELS introduces the socalled walk-ahead that takes all possible further routes into account. We conducted an extensive evaluation on a largescale real-world dataset collected from a ride-pooling service. Our experiments confirm that WHEELS reliably outperforms existing baselines and achieves an average performance gain of 19.4% compared to the best-performing baseline.