BYD (比亚迪) upgrades "Tianshen Eye" driver‑assistance with four major enhancements — architecture, sensors, algorithms and data

Upgrade summary

BYD (比亚迪) onstage in Shenzhen unveiled a major refresh to its "Tianshen Eye" (天神之眼) vehicle‑level driver‑assistance system, saying the program now includes four broad upgrades covering architecture, sensors, algorithms and data. It has been reported that the company also confirmed the Tianshen Eye B laser version will be available as an option across the lineup for ¥12,000. The announcement positions BYD to push higher‑capability ADAS and prepare for L3/L4 commercialization.

Architecture and sensor improvements

The system’s foundation is now the "Xuanji Architecture 2.0" — BYD says this uses a full‑stack, in‑house central brain and a combined cabin/drive electronics design, trimming system latency to 8 microseconds and raising bandwidth, compute and detection ranges substantially. BYD described a new "sensor satellite architecture" that uploads raw sensor signals to the central brain to reduce intermediate losses; company figures claimed a 60× increase in data bandwidth, 12× compute, 33% longer detection range and a 171% jump in diagnostic coverage. Sensor upgrades were detailed too: lidar angular resolution up 1.6× and data volume 3×, 4D millimeter‑wave radar range to 400 m with 6,000 points per frame, and ultrasound latency cut 30% with up to 10× denser point clouds.

Algorithms and data

On the software side BYD introduced a "physical AI large model" that it says fuses camera, lidar, radar and navigation inputs to "see, predict and act" — reportedly combining cloud‑based world models and long‑tail scenario simulation with on‑vehicle inference and reinforcement learning for decision optimization. It has been reported that BYD’s fleet of vehicles with assisted driving reached roughly 3.15 million units and that the system generates about 200 million kilometres of driving data per day (statistic as of May 28). BYD framed that data as the core feedback loop for continuous algorithm iteration.

L3/L4 platform and wider context

BYD also unveiled a dedicated L3/L4 platform that it says ships with three industry‑first sensors — a >1,000‑line lidar, a high‑speed "flash" camera for fast moving objects and dual infrared cameras for low‑light and fog — plus a "ten‑fold redundancy" safety architecture spanning sensors, SOCs, MCUs, power, braking, steering and communications. Observers note that Chinese automakers increasingly favor full‑stack, self‑developed solutions as they scale autonomy; against a backdrop of Western export controls on advanced chips and software, domestic vertical integration has taken on strategic importance. Will this be enough to accelerate commercial L3/L4 deployment? BYD clearly thinks so — and it has been reported that the company is backing that view with significant engineering and data resources.