Payne: Despite Musk’s claims, my Tesla unlikely to drive itself soon

Posted by Talbot Payne on April 24, 2019

On Tesla's latest, "Navigate on Autopilot" upgrade, the Tesla Model 3 will automatically change lanes on the way to its destination. That includes changing lanes when it is balked by a slower car. In this situation on the Lodge freeway, the car informed the driver of the lane change, waited for a faster car to come by, then made the pass.

On Tesla’s latest, “Navigate on Autopilot” upgrade, the Tesla Model 3 will automatically change lanes on the way to its destination. That includes changing lanes when it is balked by a slower car. In this situation on the Lodge freeway, the car informed the driver of the lane change, waited for a faster car to come by, then made the pass. (Photo: Henry Payne, The Detroit News)

Tesla laid out its road map for an autonomous future Monday, and I’m a beta tester.

In a live-streamed conference with investors, the Silicon Valley-based automaker says that cars going into production now will be equipped with the computer chip necessary to be fully self-driving. By sometime next year, predicted showman-and-chief Elon Musk, the company will also roll out a self-driving “robo taxi” service similar to Waymo and Uber that will hustle customers to their destinations — but with no one at the wheel.

Musk also boldly claimed that, after just three years of development, Tesla had developed “the best chip in the world,” which he said was seven times faster than industry’s competitors. What’s more, he was dismissive of the lidar navigation hardware that has been a cornerstone of the rest of the self-driving industry. Lidar uses pulsed lasers to allow a car to “see” what’s around it.

“Lidar is a fool’s errand,” he laughed. “Anyone relying on lidar is doomed.”

Detroit News auto critic Henry Payne took delivery on his Tesla Model 3 in October, 2018 for $57,500. To upgrade to Tesla's latest computer chip, he would need to pay an additional $5000.

Detroit News auto critic Henry Payne took delivery on his Tesla Model 3 in October, 2018 for $57,500. To upgrade to Tesla’s latest computer chip, he would need to pay an additional $5000. (Photo: Henry Payne, The Detroit News)

Tesla’s announcement came at the same time Tesla Model 3 owners like me have been receiving over-the-air updates to their autonomous “Autopilot” systems that increase the car’s ability to drive itself.

Called Navigate on Autopilot, the upgrade is a long way from self-driving, but gives owners a glimpse of the future. And it gives Tesla valuable, real-world data to test their technology.

Fun as it is, it seems generations away from full self-driving.

My state-of-the-art, Level 2 system requires constant attention from the pilot and makes for some nervy moments at the envelope’s edge. But with tens of thousands of beta testers on the road like me logging millions of miles, Musk touted Tesla’s leadership role in the race to autonomy and insisted full autonomy is right around the corner.

“The powerful sustainable force for us is the fleet,” said Musk. “We have 100 times the number of cars on the road (compared to competitors). It’s a massive data advantage.”

That plan has attracted skeptics — and billions in capital from investors such as Transpire Ventures’ John Meyer.

“Tesla’s autonomous vehicle training data clobbers anyone else aside from Waymo,” says Meyer. “This is due to the hundreds of thousands of cars that have been on the road for years, all of which have been collecting data on real-time driving … and sending it back up to Tesla.”

 
Tesla's Navigate on Autopilot software is voluntary. The driver can choose the settings they prefer from DISABLED to MAD MAX.
Tesla’s Navigate on Autopilot software is voluntary. The driver can choose the settings they prefer from DISABLED to MAD MAX. (Photo: Henry Payne, The Detroit News)

My Navigate on Autopilot came over the air (along with a 5% power boost to my already capable 271-horse system) on April 9 at midnight, just like an update to my smartphone.

The following morning, the Model 3’s giant console screen told me the car was now capable of changing lanes on its own once I had set a navigation point. Tesla allows me to set the test parameters —  from Disable to Mild to Average to Mad Max (Tesla-speak for the most aggressive setting).

Naturally, I went for Mad Max.

Like Cadillac’s SuperCruise, Navigate on Autopilot is geo-fenced — that is, it only works on limited-access highways and freeways like the Lodge and Interstate 75. As I navigated M-10 to downtown Detroit from Oakland County, the car took over from me, following the 70 mph speed limit in the slow lane. When I encountered a slower car the screen informed me it was “Changing lanes into faster lane.”

The car paused. Waited for faster cars to pass. Then turned into the left lane, accelerated around the passed car, and pulled back into the right lane. Impressive.

But my eyes never left the road. There are infinite variables that a robot car can still not see — like potholes, which are ubiquitous in Detroit this time of year. Or the odd ladder that’s fallen off a truck.

Were any of these objects in the road, I would have had to seize the wheel to avoid them — an action that the car would initially resist with a tug (hey, I’m driving!).

A fellow Tesla owner friend has no interest in wrestling with such ambiguities and tells me he’s passing on being a beta tester, thank you very much.

“That scares me,” he says.

It hasn’t scared some people enough, and there have been some terrible accidents as Tesla drivers took Autopilot at its word.

Nevertheless, Tesla insists it can overcome such obstacles with the new computer chip, part of Tesla’s so-called HW3 package that includes a radar, cameras and 12 sonar devices to navigate its landscape.

“The core problem of neural networks (the brain of a self-driving car) is recognizing objects,” said Tesla director of artificial intelligence, Andrej Karpathy. “Neural networks work better with more data.”

So when my car encounters, say, another car moving into its lane — what Tesla calls a “cut-in” — the neural network takes that experience, coupled with thousands of others, to teach cars how to drive.

“We ask the fleet to automatically send us car cut-ins,” says Karpathy. “Then we train the network. Why is Tesla unique? Because we have the fleet.”

Tesla is unique in not adding lidar to help with the task of identifying objects. That move defies rivals like Waymo, Uber and General Motors’ Cruise Automation autonomous programs.

I have driven in Waymo and Uber self-driving cars, and they are heavily reliant on lidar. Those companies, too, have promised fully self-driving cars on the road by the end of this year.

For all my car’s contributions to Tesla’s learning, I would need to upgrade its computer chip to the HW3 standard — at a cost of $7,000 on top of the $57,500 I have already spent — to be eligible for Musk’s robo-taxi service.

“People will be able to add (their own cars) to the Tesla network with Tesla taking 25-30% of the revenue,” says Musk.

My car is currently equipped with HW2 hardware which will not be able to run the software the company says will be necessary to negotiate everything that Michigan roads will throw at me: potholes, ice, blowing snow.

Call me skeptical. I doubt full self-driving is a year away, much less 10 years away.

But in the meantime I’ll be beta-testing Tesla’s software upgrades and setting them on Mad Max.

Last year in San Francisco, I took a Tesla Model S in self-driving mode onto a highway cloverleaf. The car couldn’t make it.

At the end of my trip this month when I (automatically) exited the Lodge in my Model 3, it slowed down to 35 mph and gingerly made its way around the cloverleaf.

Mission accomplished. And I was hovering nervously over the steering wheel the whole way.

Comments are closed.