Based on mass-produced 4D millimeter-wave radars and camera sensors, this hardware and software system leverages the abundant AI computing power provided by Axera's M76H chip to achieve LiDAR-like dense point cloud output, delivering higher-precision perception capability for automotive ADAS and other application scenarios. For instance, it can provide accurate perception input for the planning and control system in extreme driving scenarios such as jaywalking, heavy rain and snow, and night fog, achieving an all-round improvement in system robustness.

It boasts a maximum detection range of up to 350m, supports a moving target speed detection range of -260 to 260km/h, a horizontal field of view (HFOV) of 120°, and a vertical field of view (VFOV) of 40°. These outstanding performance parameters provide all-dimensional protection for the safety of assisted driving.

The radar signal processing and target perception tasks of this hardware are integrated into Axera's M76H domain controller computing platform, which runs more complex algorithm models to enhance system performance. The chip and its toolchain are fully supported by Axera.

Currently, the adoption of LiDAR in vehicles is a major trend, aiming to provide an extra layer of safety for intelligent vehicles. While this is the current trend, it does not mean it is the only solution. The AI radar system offers a new approach, achieving performance comparable to LiDAR with a significant reduction in cost.

As the computing core of this system, Axera's M76H chip delivers outstanding performance. Especially its self-developed core IPs, NPU and ISP, can provide 60TOPS of computing power support for the AI 4D Sensor. Meanwhile, the ISP enhances the quality of perception data of the entire system, achieving an all-round improvement in fusion effects.

The M76H is a high-computing-power automotive chip independently developed by Axera, integrated with Axera's self-developed NPU V3 architecture and featuring a heterogeneous multi-core design. It supports low-bit-width mixed precision computing: the high-efficiency tensor core is optimized for CNN and accelerates Transformer and BEV applications; the flexible vector core is tailored for Transformer optimization with flexible programmability. Complemented by SDMA and heterogeneous multi-core hardware collaboration, as well as a rich set of CV operators for pre and post-processing, it comprehensively improves perception performance.

Going forward, Axera will collaborate with FVIDAR to conduct innovative R&D based on more of Axera's automotive chips, enabling AI performance to fully empower various types of perception and providing computing-driven solutions for more industries and innovative products.