Motor-Vehicle Trajectory Prediction Task

  • Definition:The behavior of vehicles around an autonomous vehicle is uncertain. Motor-vehicle trajectory prediction aims to predict future trajectories using the historical motion trajectories of other road vehicles and map information.

  • Impact:Inaccurate trajectory prediction may cause delayed or erroneous responses from autonomous driving systems, increasing the risk of traffic accidents. Especially in complex and changing traffic environments, accurate trajectory prediction can help autonomous vehicles plan their paths in advance, reduce unnecessary emergency braking or evasive maneuvers, and avoid potential collision risks. It can significantly improve the safety and reliability of autonomous driving while enhancing the passenger experience.

  • Inputs and Outputs:Typically, motor vehicle trajectory prediction tasks take as input the recent state information (1 s or 3 s) of surrounding vehicles—possibly including coordinates, speed, and heading—as well as map information, which may include lanes, signs, markings, and traffic lights; they output vehicle state information for a future period (3 s or 5 s), such as trajectory coordinates or distributions.

Available Datasets

Dataset Details

  • NGSIM

The NGSIM (Next Generation Simulation) dataset consists of U.S. highway traffic data collected by the Federal Highway Administration (FHWA). It includes the driving conditions of all vehicles on roads such as US-101 during peak hours. The data are sampled at 5 Hz. Link:https://datahub.transportation.gov/stories/s/Next-Generation-Simulation-NGSIM-Open-Data/i5zb-xe34/

  • HighD

The HighD dataset was released by the Institute for Automotive Engineering at RWTH Aachen University, Germany. It contains real vehicle trajectory data collected on German highways, including high-precision position, speed, and acceleration information for numerous vehicles, including cars and trucks, at six locations. The data are sampled at 10 Hz. Link:https://levelxdata.com/highd-dataset/

  • Waymo

The Waymo dataset is an autonomous-vehicle dataset released by Waymo. Waymo Motion contains trajectory information for nearby motor vehicles, non-motor vehicles, and pedestrians while autonomous vehicles travel on urban roads, highways, rural roads, and other areas. The data are sampled at 10 Hz, with each segment averaging 20 s. Link:https://waymo.com/open

  • nuScenes

The nuScenes data come from 1,000 urban-road scenes in Boston and Singapore and record the trajectories of nearby motor vehicles, non-motor vehicles, and pedestrians. The dataset also provides high-definition maps. The data are sampled at 2 Hz, with each segment averaging 6 s. Link:https://www.nuscenes.org/nuscenes#overview

  • INTERACTION

The INTERACTION dataset is an international dataset created by UC Berkeley's Mechanical Systems Control Laboratory and collaborators from other organizations. It contains extensive interactive behavior by road users, such as vehicles and pedestrians, in complex intersection and roundabout scenarios in different countries. The data are sampled at 10 Hz. Link:https://interaction-dataset.com/

  • Argoverse

The Argoverse dataset was released by Argo AI. It includes more than 1,000 hours of autonomous-vehicle driving data from urban roads in Miami and Pittsburgh. The data are sampled at 10 Hz, and each scene is 5 seconds long. Link:https://www.argoverse.org/data.html#download-link