Musk Claims Tesla's New Autopilot Chip Is Only Six Months Away

Matt Posky
by Matt Posky

Tesla CEO Elon Musk says a new chip aimed at improving its vehicles’ Autopilot features will be available in about six months.

However, if you’re hoping the automaker is preparing to light some candles and knock its vehicles up with legitimate self-driving technology, you’ll need to keep on wishing. During a string of tweets on Tuesday, Musk explained that the new chip would be a $5,000 extra for customers who did not purchase their cars with the “Full Self-Driving” package — an automotive claim that’s about as valid as Donald Trump’s hair or Elizabeth Warren’s status as a Native American.

It’s difficult to understand why anyone would purchase this item for $5,000 when it’s unclear what it would even do. Musk claims it drastically improves Autopilot’s processing capabilities, boosting the maximum number of operations per second between 500 and 2,000 percent. While incredible, Tesla’s cars are still left with the same array of sensors they’ve always had.

Relying primarily on visual cues via an array of cameras, Tesla’s setup lacks some of the equipment the industry has informally agreed upon as being absolutely necessary. Tesla doesn’t use lidar and, despite Autopilot being one of the more useful and least obtrusive semi-autonomous driving aids, it’s an older system deployed somewhat irresponsibly by the manufacturer.

This new chip won’t change any of that. While we’re sure it’s theoretically possible to make the vehicle’s CPU so powerful that it can make sense of just about anything picked up by cameras, there’s a sense that we aren’t quite there yet. Otherwise, we’d probably have bipedal robots delivering our mail and sweeping our floors before they eventually rise up to destroy us.

Tesla is simultaneous promising something and nothing with this new chip. It’s reminiscent of its last over-the-air update, which included the Version 9.0 software upgrade. Musk said 9.0 effectively improved ops per second by around 400 percent after correcting an Electrek article where a “deep learning expert” over-estimated its performance to a wild degree.

However, that article also explained why something like the chip could be useful. It might not bring about an immediate evolution of the Autopilot that’s currently in your Model S, but it does help to serve the greater good. By increasing processing power on a large number of its vehicles, Tesla should also be able to transmit more useful data into the cloud. While the ins and outs of the company’s machine learning setup are still kind of murky, we know it has something in place that it uses to aid in autonomous development. This could help expedite things, resulting in new functions at a later date.

There’s also a chance it could help Autopilot function a little more smoothly in the interim, assuming its shortcoming are on the processing end of things and not down to the limitations of its sensors. Still, we’re not convinced it’s worth shelling out five grand for something that doesn’t come with any kind of immediate, tangible gratification.

[Image: Tesla]

Matt Posky
Matt Posky

Consumer advocate tracking industry trends and regulations. Before joining TTAC, Matt spent a decade working for marketing and research firms based in NYC. Clients included several of the world’s largest automakers, global tire brands, and aftermarket part suppliers. Dissatisfied, he pivoted to writing about cars. Since then, he has become an ardent supporter of the right-to-repair movement, been interviewed about the automotive sector by national broadcasts, participated in a few amateur rallying events, and driven more rental cars than anyone ever should. Handy with a wrench, Matt grew up surrounded by Detroit auto workers and learned to drive by twelve. A contrarian, Matt claims to prefer understeer and motorcycles.

More by Matt Posky

Comments
Join the conversation
2 of 23 comments
  • Sckid213 Sckid213 on Oct 17, 2018

    So it can see fire trucks now?

  • WildcatMatt WildcatMatt on Oct 18, 2018

    Matt, are you trying to channel Jack Baruth in your prose style here? That's the vibe I'm getting and pardon me for saying so but it wears like an ill-tailored suit.

  • Gray Here in Washington state they want to pass a law dictating what tires you can buy or not. They want to push economy tires in a northern state full of rain and snow. Everything in my driveway wears all terrains. I'm not giving that up for an up to 3 percent difference.
  • 1995 SC I remember when Elon could do no wrong. Then we learned his politics and he can now do no right. And we is SpaceX always left out of his list of companies?
  • Steve Biro I’ll try one of these Tesla driverless taxis after Elon takes one to and from work each and every day for five years. Either he’ll prove to me they are safe… or he’ll be dead. Think he’ll be willing to try it?
  • Theflyersfan After the first hard frost or freeze - if the 10 day forecast looks like winter is coming - that's when the winter tires go on. You can call me a convert to the summer performance tire and winter tire car owner. I like the feel of the tires that are meant to be used in that season, and winter tires make all of the difference in snowy conditions. Plus, how many crazy expensive Porsches and Land Rovers do we see crashed out after the first snow because there's a chance that the owner still kept their summer tires on. "But...but...but I have all wheel drive!!!" Yes, so all four tires that now have zero grip can move in unison together.
  • Theflyersfan One thing the human brain can do very well (at least hopefully in most drivers) is quickly react to sudden changes in situations around them. Our eyes and brains can quickly detect another driving dangerously, a construction zone that popped up while we were at work, dense fog out of nowhere, conflicting lines and signs on some highways, kids darting out between cars, etc. All of this self driving tech has shown us that it is maybe 80% of the way there, but it's that last 20% that still scares the crap out of us. Self driving computers can have multiple cameras feeding the system constant information, but can it react in time or can it work through conflicting data - think of construction zones with lines everywhere, orange signs with new exit information by the existing green exit sign, etc. Plus, and I think it's just GM's test mules, some systems require preexisting "knowledge" of the routes taken and that's putting a lot of faith in a system that needs to be updated in real time. I think in the next 15-20 years, we'll have a basic system that can self drive along interstates and highways, but city streets and neighborhoods - the "last mile" - will still be self drive. Right now, I'd be happy with a system that can safely navigate the slog of rush hour and not require human input (tapping the wheel for example) to keep the system active.
Next