Trajectory Prediction for Autonomous Vehicles in Construction Zones
Session Number
CMPS(ai) 06
Advisor(s)
Dr. John M. Dolan and Juan Alvarez-Padilla,Carnegie Mellon University
Discipline
Computer Science
Start Date
17-4-2025 2:45 PM
End Date
17-4-2025 3:00 PM
Abstract
Autonomous driving technologies have the potential to revolutionize transportation, reducing accidents, improving traffic efficiency, decreasing fuel consumption, and increasing mobility. An autonomous vehicle’s ability to generate a safe future trajectory is dependent on the outputs of many subsystems such as environment perception, intention prediction, trajectory prediction, and finally, planning. However, many of these subsystems are compromised in construction zones, as these environments are complex, cluttered, and often highly uncertain. Traffic laws do not entirely apply, road markings may be missing or misleading, lanes may have varying widths, and unknown objects may be found. This study filters the construction zones in the nuScenes dataset for the purpose of improving the AutoBots trajectory prediction architecture. The encoder architecture of the AutoBot-Ego model was modified to account for the objects in construction zones. The improved model uses nuScenes object annotations to encode theconstruction scene objects.
Trajectory Prediction for Autonomous Vehicles in Construction Zones
Autonomous driving technologies have the potential to revolutionize transportation, reducing accidents, improving traffic efficiency, decreasing fuel consumption, and increasing mobility. An autonomous vehicle’s ability to generate a safe future trajectory is dependent on the outputs of many subsystems such as environment perception, intention prediction, trajectory prediction, and finally, planning. However, many of these subsystems are compromised in construction zones, as these environments are complex, cluttered, and often highly uncertain. Traffic laws do not entirely apply, road markings may be missing or misleading, lanes may have varying widths, and unknown objects may be found. This study filters the construction zones in the nuScenes dataset for the purpose of improving the AutoBots trajectory prediction architecture. The encoder architecture of the AutoBot-Ego model was modified to account for the objects in construction zones. The improved model uses nuScenes object annotations to encode theconstruction scene objects.