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

As autonomous driving systems mature, motion forecasting has received\nincreasing attention as a critical requirement for planning. Of particular\nimportance are interactive situations such as merges, unprotected turns, etc.,\nwhere predicting individual object motion is not sufficient. Joint predictions\nof multiple objects are required for effective route planning. There has been a\ncritical need for high-quality motion data that is rich in both interactions\nand annotation to develop motion planning models. In this work, we introduce\nthe most diverse interactive motion dataset to our knowledge, and provide\nspecific labels for interacting objects suitable for developing joint\nprediction models. With over 100,000 scenes, each 20 seconds long at 10 Hz, our\nnew dataset contains more than 570 hours of unique data over 1750 km of\nroadways. It was collected by mining for interesting interactions between\nvehicles, pedestrians, and cyclists across six cities within the United States.\nWe use a high-accuracy 3D auto-labeling system to generate high quality 3D\nbounding boxes for each road agent, and provide corresponding high definition\n3D maps for each scene. Furthermore, we introduce a new set of metrics that\nprovides a comprehensive evaluation of both single agent and joint agent\ninteraction motion forecasting models. Finally, we provide strong baseline\nmodels for individual-agent prediction and joint-prediction. We hope that this\nnew large-scale interactive motion dataset will provide new opportunities for\nadvancing motion forecasting models.\n