Our Perception module combines the strengths of a heuristic approach and deep learning models to boost performance, while ensuring the safety and operational redundancy of our vehicles. Performance capability is further enhanced by our multi-sensor fusion technology, which intelligently leverages the reliable sensor data depending on different environmental or driving scenarios.

The Prediction module works to project how other vehicles, pedestrians, and objects (together, “road agents”) may move or behave based on several inputs, including Perception output, raw sensor data, and data regarding previous decisions made by the road agent (i.e. sequence of events). The output of the Prediction module for a certain road agent is a series of predicted trajectories, each with an assigned probability of occurrence.

We fuse machine learning and deep learning to create a robust Planning and Control module able to smoothly navigate complex road scenarios — from the streets and highways of California to the bustling eight-lane intersections of Chinese metropolises, while being prepared for outlier behaviors or events caused by other road agents.

Successful autonomous driving technology deployment and scale rely on an entire set of supporting infrastructure. From the real-time onboard system to off-vehicle data processing and management, we have built a solid foundation that drives rapid iteration, scalable development of a high-quality system, and efficient testing in all aspects of software and hardware development.

Excellent performance, high reliability, and deployment-ready are the key guiding principles for our hardware solutions. Our engineers custom design the sensor fusion module and compute system to bring high performance components together with cutting-edge software, resulting in a tightly integrated full-stack system. All driving and safety-critical elements are equipped with hardware redundancies.