People Who Worked On Tesla “Full Self Driving” Don’t Trust Tesla “Full Self Driving”

Support CleanTechnica’s work through a Substack subscription or on Stripe.

Some of them, at least.

According to Reuters, the news organization interviewed a former self-driving engineer at Tesla and nine former data labelers, and they did not offer a singing endorsement of Tesla’s “Full Self Driving” (FSD) system. Reuters also talked with eleven independent traffic safety researchers for the piece.

“In a Utah office, hundreds of Tesla workers scrutinize video collected by vehicles using the automaker’s Full Self-Driving (FSD) feature. Some clips show the cars hitting cats, dogs or deer, along with more-routine accidents. Sometimes, they don’t brake before impact,” the piece starts out. “Often, they speed. Occasionally, the workers see near-misses of children playing in the street.” It doesn’t get much better from there.

The former staff countered one of the most popular claims from Tesla fans and Elon Musk — that true unsupervised FSD could just be turned on broadly, across the US at least, soon and every Tesla driver with FSD could do other things while the cars drove themselves. Counter to that, the former staff said they worked for hours training FSD on specific routes to make the system good enough to show off for public events, and that deploying FSD broadly without all of that focused work could not be done — that the system simply wasn’t good enough. “The staffers said these labor-intensive safeguards are impossible to deploy on a broad scale,” as Reuters summarized it. “Those efforts, which haven’t been previously reported, undermine Musk’s long-stated claim that Tesla’s self-driving technology will soon work anywhere globally and doesn’t require the same laborious local mapping of roads and hazards employed by rivals.”

For many years — about a decade actually — Tesla fans and Elon Musk were super critical of Waymo and other approaches that needed to train and deploy in geofenced areas. The idea was that while other systems were limited by this need for geofencing, Tesla’s system was general and once it got good enough, Tesla could “flip a switch” and turn on the capability everywhere. As Tesla has started deploying early-stage robotaxis, however, the company has geofenced them and has limited them to a very small fleet of cars. Of course, the idea from fans is still that one day Tesla will have FSD at a good enough level that it will be able to flip that switch. But the story so far, including from these former staffers, is that Tesla needs concerted, localized work and geofencing.

Out of those nine former data labelers, seven said they wouldn’t trust FSD to drive them. One even said they wouldn’t ride in a Tesla robotaxi even if someone paid him to do so. Yikes!

“The safety statistics Tesla uses publicly are equally shaky. Researchers found that Tesla inflates its safety numbers by comparing airbag-deployment crashes in FSD vehicles against a federal crash database that includes far less severe incidents, and by benchmarking against the average American car, which is considerably older than the average Tesla,” Autoblog summarizes. “Ten of the eleven researchers who reviewed the methodology told Reuters it read more like marketing than genuine safety analysis.”

These critiques are not really new, but the claims of these former staffers are about as stark as you can find, and they saw the process and problems close up. Also, regarding the safety statistics, we’ve pointed out repeatedly that comparing Teslas using FSD or Autopilot to all cars broadly has that massive problem with regard to age of vehicle as well as where those miles are being racked up, but I didn’t realize there was that issue with the different types of crashes. Tesla was comparing only airbag-deployment crashes with all crashes? That’s nuts. That’s even considerably more unbalanced and deceitful than I expected.