The Prediction team liaises with the perception team upstream and the planning and control team downstream. We develop diverse deep learning models using perceptual information to forecast potential future behaviors of nearby obstacles. In scenarios lacking sufficient data, the prediction team generates predictions and supplements through analysis of scene features and logical reasoning.

The main goal of the planning and control team is to find safe driving strategies in complex autonomous driving environments. We ensure safe, efficient navigation by designing algorithms for route planning and vehicle control to respond to dynamic conditions. Our work integrates sensor data and predictive models to make real-time decisions for smooth and reliable autonomous driving.

The mapping and localization team is responsible for high-definition mapping and high-precision localization systems for our entire autonomous fleet. We provides down-stream modules, e.g. perception, prediction, planning and control, with accurate and real-time positioning as well as detailed geometric and semantic information of the surroundings, to help the autonomous vehicle navigate the world. A variety of underlying technologies are employed to build the mapping and localization systems, including multi-sensor fusion, visual/lidar SLAM, 3D computer vision and deep learning.

The Sim Core group focuses on developing the core simulation capabilities necessary to support company-wide AV development and validation. This team will work on our overall simulation strategies, sensor simulation for perception, AI simulation for prediction, planning and control, full-stack simulation, and the generation and editing of 3D worlds.