Lidar was once the most contested sensor in autonomous driving. Supporters argued it supplies 3D information cameras cannot, a safety redundancy high-level driver assistance cannot do without. Opponents, led by Tesla, called it expensive, bulky and transitional. Yet since 2025, as advanced driver assistance spread, lidar has not exited, it has entered mass production, and humanoid robots, delivery vehicles and industrial automation have become important new customers.
One chip, many scenes
As chip-based architectures mature, the SPAD-SoC approach shows cross-scene reach, and the industry is moving from car-only to a ‘car plus robot’ dual drive, covering multiple scenarios with one underlying chip. A newer lidar vendor puts it plainly: a car and a robot are not fundamentally different, the core difference is in production control and test standards, not the chip itself. RoboSense’s ‘Genesis’ platform, launched in April, is a SPAD-SoC digital base that builds a narrow-array ‘Phoenix’ chip for high-line car use and a large-array ‘Peacock’ chip for robots and blind-spot radar, both heading to volume production this year. Its earlier large-array SPAD-SoC chip already ships in the E1 and E1R products, used for ADAS, Robotaxi and humanoid robots, with more than 300,000 units sold. Hesai, too, is extending from vehicles into consumer spatial intelligence and robot core parts.
The RGB-D endgame
The long-standing split between lidar and camera creates spatial and timing alignment problems that fusion software never fully fixes, and two hardware sets raise cost. The ultimate form is single-chip, single-lens RGB-D, generating colour and depth together at the pixel level, at a cost close to an ordinary camera. Hesai’s ‘Picasso’ SPAD-SoC is a 6D colour lidar chip that fuses 3D perception and colour on the chip and outputs a colour point cloud; its ETX series supports up to 4,320-line full-colour 4K sensing and is planned for volume delivery in the second half of the year. The catch is resolution: today’s single-chip RGB-D sits near 1 megapixel or lower, while car cameras start at 8 megapixels, so it needs to clear 4 megapixels before the market takes it seriously.
From cost to chip design
Chipisation is reshaping the contest. In the past the fight was consolidated production and cost; now it is chip design ability. Swap one spec of chip on the same base and you move from car to robot to industrial sensing to drone. As heads such as Huawei, RoboSense and Hesai pushed line counts to 896 and 2,160 and range past 600 metres, a high-line arms race formed, but the real divergence is opening between full-solid-state large-array SPAD-SoC and the older rotating-mirror approach, two generations of technology.
Read the original report (LeiPhone)
Translated and adapted from LeiPhone (leiphone.com).