Ex-Tesla Autopilot head calls vision-only a 'weird religious stance'

Sterling Anderson, GM’s chief product officer and former leader of Tesla’s Autopilot program, argues that relying only on cameras for self-driving is an engineering mistake because cameras fail in conditions such as dense fog and glare and cannot directly measure speed as radar can. He says radar is inexpensive, while lidar can also help camera systems improve when their data are used together. Anderson’s criticism comes as Tesla deploys camera-only Cybercabs in Austin. GM is pursuing a different approach: its planned 2028 Cadillac Escalade IQ “eyes-off” system is expected to combine lidar, radar, and cameras. Anderson also cautions against “eyes-off” marketing, saying highly capable driver assistance can encourage drivers to stop paying attention.

The piece supports Anderson’s case with Tesla’s history of removing radar from Model 3 and Model Y vehicles in 2021 and ultrasonic sensors in 2022, despite reported objections from company engineers. It highlights a federal investigation covering about 3.2 million vehicles, expanded after nine crashes, including a pedestrian fatality. NHTSA found that Tesla’s system failed to detect visibility problems such as glare, fog, and dust, sometimes warning drivers only shortly before crashes. The central argument is that sensor redundancy matters when cameras cannot see reliably. It also contrasts Tesla’s roughly three million driverless miles with Waymo’s reported 271 million rider-only miles, while noting that Tesla now uses geofenced service areas and is building fleet maps despite previously rejecting such approaches.