GM Chief Product Officer Sterling Anderson, who ran Tesla’s Autopilot program in 2015 and 2016, says dropping radar in favor of a camera-only self-driving system is “a weird religious stance that ignores physics”.
It’s the bluntest rebuke of Tesla’s vision-only bet yet from someone who helped build the original Autopilot. And it lands while Tesla is pushing camera-only Cybercabs into driverless service in Austin.
Anderson made the comments in an interview with The Verge a week ago after GM’s technology event in New York the week before. Asked whether GM could eventually move to a vision-only approach like Tesla’s, he left little room for interpretation.
“If you scale back to vision-only, physics suggests you’ll be in trouble in certain situations,” Anderson said. He added that radar is so cheap that removing it makes no sense, and that “cameras can’t see through dense fog or measure instantaneous velocity like radar can. Having a multisensor suite provides obvious safety advantages that physics simply demands.”
He also said lidar (Light Detection and Ranging) helps camera systems improve faster when the two are trained together. That directly counters Tesla’s argument that cameras plus fleet data are all you need.
Automotive veteran
Sterling Anderson isn’t a lidar startup CEO talking his book. He led the Model X program at Tesla, then directed Autopilot during its formative years before co-founding self-driving company Aurora. GM hired him as chief product officer (CPO) in 2025, and Bloomberg reported last year that Mary Barra sees him as a frontrunner to become GM’s next CEO.
His roadmap at GM runs opposite to Tesla’s. The Cadillac Escalade IQ is due to debut GM’s “eyes-off” system in 2028, with lidar, radar, and cameras.
He took a shot at another Tesla habit in the same interview. Anderson said he avoids “eyes-off” marketing language altogether, arguing that a driver-assist system that feels very capable already pushes drivers into complacency. That’s the core problem with Tesla selling ‘Full Self-Driving (Supervised)’.
Visibility is key
Tesla started removing radar from Model 3 and Model Y in May 2021. The Tesla Vision system launched with limitations: Autosteer capped at 75 mph, a longer minimum following distance, and Smart Summon and Emergency Lane Departure Avoidance disabled at delivery. Ultrasonic sensors were eliminated in 2022.
The decision wasn’t unanimous inside Tesla. Former employees said engineers tried to convince Elon Musk not to drop radar, and he overruled them. Musk told the specialized site Electrek at the time that “a very high resolution radar would be better than pure vision, but such a radar does not exist”. Electrek argued then that plenty of high-resolution radars exist today.
But the bigger problem is the one Anderson is pointing at: visibility. In March, NHTSA upgraded its investigation into Tesla’s FSD system in reduced-visibility conditions to an Engineering Analysis covering 3,203,754 vehicles, the last step before a recall.
The probe now counts nine crashes, including a pedestrian killed in November 2023. NHTSA found FSD’s camera degradation detection “did not detect common roadway conditions that impaired camera visibility,” including sun glare, fog, and airborne dust, and only alerted drivers right before the crashes. Radar doesn’t care about sun glare. That’s the point of redundancy.
Anderson has the better arguments, and he’s in a better position than almost anyone to say it. He ran Autopilot back when it still used radar.
The vision-only case
Electrek has his doubts about the Tesla approach. “The vision-only case always rested on a tidy idea: humans drive with two eyes, so cameras and enough AI should do the job. But humans crash in fog and glare too, and the promise was always to be far safer than humans, not just as good.”
“Removing a cheap sensor that directly measures velocity, overruling your own engineers to do it, and ending up with a 3.2-million-vehicle NHTSA investigation into cameras that can’t tell when they’re blind isn’t first-principles engineering. It’s a position Musk staked out in 2019 when he called lidar a “fool’s errand,” and he’s been defending it ever since,” Electrek signals.
“Tesla’s own behavior has drifted anyway. It’s building its own fleet maps after years of mocking them, and its Robotaxi runs inside geofences like everyone else’s. Meanwhile, Waymo, with lidar, radar, and cameras, has logged roughly 271 million rider-only miles. Tesla has about 3 million driverless miles since it started pulling safety monitors in January,” Electrek concludes.
