Data-DrivenData-driven algorithms in full process with less cost and higher effciency vs.
Massive DataCovering real scenarios and corner cases with massive data.
Closed Loop AutomationHighly automated and efficient toolchain to process the massive data
and solve the problem at low cost.
CLA enables us to quickly and cost-efficiently process, filter and label massive amounts of data and deploy our algorithms against more applications and driving scenarios. This drives our algorithms to become smarter and more sophisticated, leading to efficient iteration and improved performance of our technology and solutions which in turn generates more data that we can use to refine our algorithms.
Our perception algorithms transform sensory information into actionable information about the vehicle and its environment in real time. Over the years, our world-leading experts of deep-learning have built proprietary algorithms which achieve high detection accuracy, with rich features, an engineering-friendly architecture when running in vehicles.
Human keypoint detection
human pose and adction recognition
3D positioning and motion prediction
1000+ vehicle models recognition
Bounding box in 3D space
Lanes and guide lines at intersections
Road signs detection and recognition
Traffic light recognition
Free space recognition
More complicated VRU behaviors
Different traffic sign and boundaries
Unique vehicles in China
HD Map for autonomous driving entails features of precision, freshness and scalability. 3D positioning of the surrounding environment is reconstructed by extracting semantic information from 2D images and fusing with GPS and IMU data. “Live” map which powers various levels of autonomous driving is automatically generated from vehicles with low-cost and mass-produced sensors. HD Map is used by different autonomous driving software modules such as localization, planning and control.
Data-Driven Path Planning
Leveraging our [proprietary] AI deep learning technologies, our planning algorithms guide vehicles through a variety of driving moves. Under our data-driven approach, our planning algorithms efficiently adapt to traffic rules or driving styles in new scenarios compared with the more commonly used rule-based approach. This allows our planning algorithms to achieve high scalability as it avoids the large engineering effort to constantly update planning algorithm logic.