One of the main drivers of research in autonomous driving is the quest to increase road safety and reduce accidents. Highway driving not only constitutes majority of the driving time in developed nations, but is also most prone to deaths due to accidents because of the higher speeds. The objective of this project is therefore to develop an online threat assessment system for highway driving scenarios, with special focus on predicting lane change maneuvers of surrounding traffic, based solely on information available through on-board perception sensors.
Past work on the project highlighted the importance of obtaining accurate lateral velocity estimates for lane change prediction through an HMM based approach. Also, the state of the art lane change maneuver prediction approaches do not handle vehicle interaction explicitly which has been highlighted as being crucial for robust threat assessment. The aim is to better understand and model the effect of tracking noise and interactions between not just the ego vehicle and the relevant traffic but also among the vehicles constituting the relevant traffic.