Firmware

Tesla’s HW3 upgrade next year will usher in Full Self-Driving, improved Neural Net computing abilities

Tesla’s recently held Q3 2018 earnings call provided several updates on the company’s ongoing Hardware 3 development, which would help its fleet of electric cars achieve Full Self-Driving capabilities in the future. According to the company’s executives, work on the custom hardware is continuing, and prototypes of HW3 have already been installed on test units of the Model S, X, and 3.

The updates on Tesla’s Full Self-Driving plans and the development of the third version of Autopilot hardware were announced during the opening minutes of the earnings call. Pete Bannon, who heads the team developing the Tesla’s custom AI chip, noted that the company would be building a manufacturing line for HW3, emphasizing the vertical integration strategy that has become a trademark of the electric car maker. 

Bannon also stated that development on the third-generation hardware saw notable progress in the quarter. The qualification of the chip and board, for one, as well as the qualification of its manufacturing lines, were conducted in Q3. Test versions of Model S, Model X, and Model 3 vehicles equipped with Hardware 3 have also been completed to validate the homegrown hardware.

Ultimately, Tesla’s progress on its Autopilot suite comes amidst the company’s focus in the development and eventual rollout of its Full Self-Driving suite, which is expected to happen in the second quarter of 2019. During the earnings call, Elon Musk noted that HW3 should provide Tesla’s electric cars with a 1000% improvement in processing capability compared to current hardware. Musk also reiterated that the cost of HW3 would be identical to HW2, despite a tenfold improvement in performance.

“[It’s] very important to emphasize that the only thing that needs to change between a car that’s produced today and a car that’s going to be produced in the second quarter of next year is probably the autopilot computer. This is a simply change that takes about half an hour to upgrade the computer. Anyone will be able to upgrade their car to Full Self-Driving capability with a simple service visit. Basically anything made in the last two years will be upgradable to Full Self-Driving. So its better important to emphasize that there’s no need to wait until that comes out because its just a very simple plug and play change to get to Full Self-Driving. Anyone who has made for Full Self-Driving option will get it done for free. It really ends up being a discount on future capability.”

Tesla Director of Artificial Intelligence Andrej Karpathy also provided some details on the improvements of the company’s neural networks, which have been growing steadily as the size of the fleet increases. The AI director noted that the upcoming hardware would be pivotal in supporting the growing demands of Tesla’s ever-growing neural net.

“The team is incredibly excited about the upcoming upgrade for the Autopilot computer. This upgrade allows us to not just run the current neural networks faster, but more importantly, it would allow us to deploy much larger computational and more expansive networks to the fleet. The reason this is important is that it is a common finding in the industry that as you make the networks bigger, the accuracy of their prediction increases with the added capacity. Now we’re currently at a place where we’ve trained large networks that work very well, but are not able to deploy them to the fleet due to computational constraints. So all of these will change with the next iteration of the hardware, and it’s a massive step improvement in computing capability, and the team is incredibly excited to get these networks out there.”

 

As Tesla prepares for the rollout of Hardware 3, the company is the also setting the stage of the release of its most advanced iteration of Autopilot to date. During the official release of V9, Tesla opted to hold back Navigate on Autopilot, a feature that allows vehicles to automatically change lanes, handle forks in the road, and conduct on-ramp and off-ramp maneuvers. Stewart Bowers, Tesla’ VP for Engineering, described the upcoming feature in the Q3 earnings call.

“We will soon roll out the team’s most advanced autopilot ever, Navigate on Autopilot. In our last release we watched a new set of neural networks that combined together provide a view of everything happening in and out of the car. With Navigate on Autopilot, we’ll use information to understand exactly where the car is on the highway system and to automatically change lanes, handle forks, and take high-curvature exits to follow a map route. Initially it will require drivers to confirm lane changes using the turn signal before the car moves into an adjacent lane. Future versions will allow customers to waive the confirmation requirements if they choose to.

“One area that I’m personally really excited to build on in this improvement is active safety. With the advancement of neural network covering 360 views around our car we can provide a level of constant vigilance that humans just can’t. Ultimately, this will allow us to warn and even intervene for an enormous percentage of modern accidents and to ship these improvements as software upgrades to our existing customers. By bringing out more cameras around the car, we  can detect things as they come toward us not just directly in front of us.”

Initial demos of Navigate on Autopilot from members of Tesla’s early access program have teased some of the capabilities of Navigate on Autopilot, which includes Elon Musk’s fondly-named “Mad Max” setting, which determines the aggressiveness of the electric car when changing lanes and maneuvering around traffic. As could be seen in videos of the feature, though, even Tesla’s Mad Max setting is incredibly careful on the road, initiating overtaking maneuvers only when it is safe.

Tesla’s HW3 upgrade next year will usher in Full Self-Driving, improved Neural Net computing abilities
View Comments
To Top