In his recent biography of Elon Musk, Walter Isaacson revealed that Musk had a near-death experience while testing Tesla's Autopilot in 2015. The car swerved into oncoming traffic due to faded lane markings, which were not detected by the car's vision sensors. Musk was furious and demanded that his engineers fix the problem.
Tesla's engineers believed that the car needed other sensors, such as radar, to improve safety. However, Musk insisted that Tesla cars could rely solely on vision sensors, arguing that humans also use their eyes to drive. In the end, Tesla's engineers were unable to convince Musk to change his mind.
Instead, Tesla's engineers took a simple but crude approach: they repainted the lane markings on the road. This solution worked, and Autopilot was able to avoid the dangerous curve.
This incident suggests that Tesla's engineers are not opposed to using other sensors, but that Musk is reluctant to do so due to cost and technical concerns.
Chinese automakers, on the other hand, are increasingly adopting LiDAR. This is a pragmatic approach that acknowledges the limitations of vision-only systems.
As LiDAR technology continues to improve in China, Tesla's engineers are likely to become more vocal in their support of LiDAR.
Good LiDAR systems should meet three key criteria:
- High-quality point clouds: LiDAR systems should generate point clouds that are accurate, real-time, and have a wide range.
- Affordability: LiDAR systems should be affordable enough to be widely adopted.
- Reliability: LiDAR companies should have a strong track record of reliability.
China is already a major market for LiDAR, and Chinese companies are leading the way in developing advanced LiDAR technologies. As LiDAR costs continue to fall, Chinese companies are likely to play an increasingly important role in the LiDAR market.
In conclusion, Tesla's self-driving technology is evolving from "pure vision" to "vision+LiDAR." This trend is driven by the need for improved safety and performance.