By integrating domestically developed sensors and computing modules, a heterogeneous sensor fusion framework is established. An AI multi-task model is applied to develop object detection, lane marking detection, and drivable space perception. In addition, a Hidden Markov Model (HMM) is employed to predict the intent of forward objects.These capabilities are utilized to handle vehicles cutting into the lane, assist drivers during lane changes, and enable surround-view obstacle detection, thereby meeting international regulatory requirements, including: l UN R79 ESF (Emergency Steering Function): Vehicle and minimum obstacle detection requirements l UN R79 ACSF-C (Lane Change Assist Function): Left and right blind-spot long-range detection up to 70 meters UN R157 (ADAS Level 3): Cut-in vehicle detection requirements

•Input voltage range: 24 V, 280 W 

•Operating temperature range: –40 °C to 85 °C 

•Sensing modules: radar ×6 ，cameras ×6， 4D radar ×1 

•Interfaces: GMSL / Ethernet / CAN 

•Sensor specifications: 

    Camera: 720p ;  

    Radar: 77 GHz  ;  

    4D radar (or LiDAR: 32 layers)

•Detection Range: 

    Front: up to 100 m 

    Sides: up to 10 m 

    Rear: up to 75 m 

•Output information: Detected objects (pedestrians, bicycles, passenger vehicles, commercial vehicles, motorcycles)Lane marking features and third-order polynomial equations Predicted intent of objects (pedestrians and vehicles) within the next 1 second, including left/right turning, acceleration/deceleration, lane crossing, as well as position and velocity

This technology can be applied to Level 2~Level 4 self-driving cars, with sensor module synchronization and integration technology. The sensing range is applicable to meet the detect and tracking requirement for ESF and ACSF-C of UN R79.

The technology can make domestic braking component manufacturers have capabilities of system integration, and master the key technologies of synchronous integration modules, meeting the system requirements of Level 2~Level 4 autonomous vehicles.