Tesla releases FSD autonomous driving chip, diss LiDAR

At Tesla's Autonomy Investor Day in late April, founder Elon Musk made a high-profile announcement about the autonomous driving chip "full self-driving computer" (FSD), which has been installed in the latest Tesla models.

Musk is very dissatisfied with the currently commonly used LiDAR solution and clearly stated that "any company that relies on LiDAR solutions will fail", so he plans to launch his own visual sensor solution.

The biggest limitation of LiDAR is the cost. The price of a 64-beam LiDAR for high-level autonomous driving is as high as 500,000 yuan, and this cost will eventually be passed on to consumers. This may be the main reason why Musk insisted on using the visual sensor solution.

The cost of visual sensors is much lower. In essence, it is a camera that works with scientific algorithms. Although it is not as good as lidar in terms of ranging and high-maneuverability response, it is also being gradually optimized. Visual sensors also have stronger capabilities in environmental perception and spatial three-dimensional modeling.

The binocular inertial module can already continuously output 200 frames of images and IMU data. The module can quickly detect changes in road curvature, shadows, road surface and marking lines at a speed of 0.005 seconds per frame, and analyze lane change trends through continuous multi-frame analysis. It can provide the autonomous driving algorithm with multi-faceted, accurate and fast environmental information data, enabling the autonomous driving vehicle to have a high-maneuverability response to road conditions.

The FSD computing power released by Tesla this time is as high as 144 TOPS. Pete Bannon, head of Tesla's autonomous driving, said that the chip can process 2,100 frames of image input per second, equivalent to 2.5 billion pixels per second. This is the chip with the highest computing power in use in the world.

This computing power enables high-definition visual sensor images to be accurately calculated at high speed, and used to make decisions for autonomous driving and improve responsiveness. This is also the reason why Tesla dared to abandon lidar and mainly use visual sensors.

Of course, visual sensors are not without shortcomings. For example, they are affected by extreme weather conditions such as light, heavy rain, fog and snow. Therefore, it is still unrealistic to completely remove other sensing devices and use only visual sensors. Therefore, Tesla also has millimeter-wave radar on the front of the car.

Musk also revealed that Tesla is already researching the next generation of chips, which is expected to have three times the computing power of FSD. With more complete artificial intelligence algorithms, it will become more possible to replace lidar with visual sensors to reduce costs.