From swerving into a median to narrowly missing poles, videos of Tesla’s latest Full Self-Driving update don’t inspire much confidence

The interior of a Tesla driving down the highway
Tesla interior Reuters
  • Tesla launched the latest version of its Full Self-Driving beta after a long delay this month.
  • Beta testers immediately started posting clips of the driver-assistance system.
  • The software is impressive and advanced, but still gets drivers into dangerous situations.
  • See more stories on Insider’s business page.

Earlier this month, Tesla rolled out a long-awaited update to its Full Self-Driving software for beta testers. It’s impressive – but it still doesn’t make cars autonomous.

The electric-vehicle maker first beamed out access to the pre-production tech in October, and it’s now in the hands of a couple thousand loyal Tesla owners. It takes Tesla’s existing driver-assistance system, which is mainly suited for predictable highway driving, and adds the ability to automate driving tasks on more complicated non-highway streets.

Videos of the new-and-improved software in action show that it can impressively navigate some tough driving situations, but there are plenty of dangerous flaws and glitches too.

In one clip, a Tesla confidently handles a tight, unmarked road with an oncoming car. The computer does pretty much exactly what a human would do: slow down and pull over to let the oncoming car go first, then pull forward once it’s clear that the other driver is giving right of way.

-Eli Burton (@EliBurton_) July 12, 2021

Another shows the system navigating stop-and-go traffic:

-Whole Mars Catalog (@WholeMarsBlog) July 11, 2021

Another shows that it can see stop signs and make turns on dark – albeit empty – city streets, too. Some videos also show cars stopping for pedestrians and other vehicles.

Read more: 
Fort Lauderdale asked Elon Musk to build a commuter train tunnel. So how did it end up considering a $30 million beach tunnel for Teslas instead?

But the system still struggles with utterly basic driving tasks, putting drivers and bystanders in dicey situations. In one clip documenting a drive in downtown San Francisco, the car drunkenly swerves into a striped median, forcing the driver to take control.

-Manish Vij (@vijmn) July 13, 2021

In the same video, the car stumbles through a left turn and nearly oversteers into a parked car.

In a clip set in Chicago, the car slowly creeps through intersections, comes to random stops, and only notices a road closure at the last second. A bunch of orange barricades is something any average human driver would recognize before actually attempting a turn.

All of these dangerous hiccups show just how far Tesla is from replicating human driving. But one particularly alarming clip out of Seattle takes the cake.

In the nighttime video, the beta fails to recognize the massive concrete columns supporting the city’s monorail – and the car nearly steers into them twice in an attempt to change lanes.

If a highly automated car should be able to do one thing, it’s recognizing large stationary objects and avoid them. But it appears that the car had no idea the pillars were even there, judging by the visualization displayed on the center screen.

The people in the car wonder whether the failure is a result of Tesla shifting to a camera-only system that doesn’t use radar. And that’s certainly a possibility. Car companies, Tesla included, have relied on radar for years for features like emergency braking and cruise control. But Tesla in May decided to stop using the sensors and take them out of its future cars.

Tesla has adopted a fast-and-loose approach to its automated-driving tech that other automakers aren’t taking. And safety advocates have taken issue with Tesla’s strategy of having amateur drivers test unproven technology on public roads. Pedestrians, cyclists, and other drivers didn’t sign up to be subjects in this lab experiment, they argue.

But the company is under mounting pressure to deliver a final version of Full Self-Driving to customers, who have shelled out up to $10,000 over the years for the add-on under the promise that it would eventually enable Teslas to drive themselves. It’s increasingly looking like that’s not happening anytime soon.