The Waymo Open Dataset is a large-scale, high-quality, and diverse dataset designed for autonomous driving research. It consists of 1150 scenes, each spanning 20 seconds, with synchronized and calibrated high-quality LiDAR and camera data captured across various urban and suburban areas. The dataset is 15 times more diverse than the largest existing camera+LiDAR dataset based on geographical coverage. It includes 12 million LiDAR and 12 million camera annotations, providing 113,000 LiDAR object tracks and 250,000 camera image tracks. The dataset includes 2D and 3D bounding boxes for vehicles, pedestrians, cyclists, and signs, with consistent identifiers across frames. It also provides rolling shutter aware projection libraries for sensor fusion research. The dataset is used to evaluate 2D and 3D object detection and tracking methods. It includes a wide range of scenes across different cities, including San Francisco, Phoenix, and Mountain View, with a total geographical coverage of 40 km² in Phoenix and 36 km² in San Francisco and Mountain View. The dataset is designed to help researchers understand the effects of dataset size and generalization across geographies on 3D detection methods. It also includes a domain gap study, showing that models trained on data from one domain may not perform well on data from another domain, highlighting the importance of domain adaptation research. The dataset includes a variety of tasks, including 2D and 3D object detection and tracking. It provides baselines for these tasks, including 3D LiDAR detection using PointPillars and 2D object detection using Faster R-CNN. The dataset also includes multi-object tracking baselines, using methods such as Tracktor and a Kalman Filter-based approach. The dataset is publicly available, and a public leaderboard is maintained to track progress in the tasks. Future plans include adding more labeled and unlabeled data with a focus on different driving behaviors and weather conditions to enable research on other self-driving tasks such as behavior prediction and planning.The Waymo Open Dataset is a large-scale, high-quality, and diverse dataset designed for autonomous driving research. It consists of 1150 scenes, each spanning 20 seconds, with synchronized and calibrated high-quality LiDAR and camera data captured across various urban and suburban areas. The dataset is 15 times more diverse than the largest existing camera+LiDAR dataset based on geographical coverage. It includes 12 million LiDAR and 12 million camera annotations, providing 113,000 LiDAR object tracks and 250,000 camera image tracks. The dataset includes 2D and 3D bounding boxes for vehicles, pedestrians, cyclists, and signs, with consistent identifiers across frames. It also provides rolling shutter aware projection libraries for sensor fusion research. The dataset is used to evaluate 2D and 3D object detection and tracking methods. It includes a wide range of scenes across different cities, including San Francisco, Phoenix, and Mountain View, with a total geographical coverage of 40 km² in Phoenix and 36 km² in San Francisco and Mountain View. The dataset is designed to help researchers understand the effects of dataset size and generalization across geographies on 3D detection methods. It also includes a domain gap study, showing that models trained on data from one domain may not perform well on data from another domain, highlighting the importance of domain adaptation research. The dataset includes a variety of tasks, including 2D and 3D object detection and tracking. It provides baselines for these tasks, including 3D LiDAR detection using PointPillars and 2D object detection using Faster R-CNN. The dataset also includes multi-object tracking baselines, using methods such as Tracktor and a Kalman Filter-based approach. The dataset is publicly available, and a public leaderboard is maintained to track progress in the tasks. Future plans include adding more labeled and unlabeled data with a focus on different driving behaviors and weather conditions to enable research on other self-driving tasks such as behavior prediction and planning.