Large Scale Interactive Motion Forecasting for Autonomous Driving : The Waymo Open Motion Dataset

MultiPath++: Efficient Information Fusion and Trajectory Aggregation for Behavior Prediction

“…Nearly all work assumes an independent, per-agent output space, in which agent interactions cannot be explicitly captured. A few works are notable in describing joint interactions as output, either in an asymmetric [28,47] or symmetric way [18,35,41].…”

“…There has been a rich body of work on how to model agents' futures, their interactions, and the environment. However, there is little consensus to date on the best modeling choices for each component, and in popular benchmark challenge datasets [6,12,18,53], there is a surprisingly diverse set of solutions to this problem; for details see Section 2 and Table 2.…”

“…Lastly, we explore ensembling and output aggregation techniques-common in other ML domains-and find effective variants for our probabilistic multimodal output representation. We perform an extensive ablation on these design choices, and show that our proposed model achieves state-of-the-art performance on the Argoverse Motion Forecasting Competition [12] and the Waymo Open Dataset Motion Prediction Challenge [18].…”

V2X-Sim: Multi-Agent Collaborative Perception Dataset and Benchmark for Autonomous Driving

“…As for sensor modalities, nuScenes [3] collected data with Radar, RGB camera, and LiDAR in a 360 • viewpoint; WoodScape [43] captured data with fisheye cameras; and A2D2 [12] provided extensive vehicle bus data including the steering wheel angle, throttle, and braking. Regarding data annotations, semantic labels in both images [7,15,30,36] and point cloud [2,14] were provided to enable semantic segmentation; 2D/3D box trajectories were offered [4,9] to facilitate tracking and prediction. In summary, existing datasets generally emphasized the data comprehensiveness in single-vehicle situations, but ignored the multi-vehicle collaborative self-driving scenarios.…”

Vehicle trajectory prediction works, but not everywhere

“…Vehicle trajectory prediction is one of the main building blocks of a self-driving car, which forecasts how the future might unroll based on the scene, i.e., the road structure, and the traffic participants. State-of-the-art models are commonly trained and evaluated on datasets collected from few cities [14,19,23]. While their evaluation has shown impressive performance, i.e., almost no off-road prediction, their generalization to other types of possible scenes e.g., other cities, remains unknown.…”