Tesla is currently using a hardware suite of sensors from Mobileye to get the input needed for its Autopilot system to control the Model S on highways and under some specific road conditions. But as it is often the case with Tesla (like for its seats), the company is looking to discontinue Mobileye’s system in favor of bringing it in-house, according to an email exchange between Tesla CEO Elon Musk and George Hotz, a software engineer mainly known for being the first person to jailbreak the iPhone.
George Hotz, also known as Geohot, hacked the iPhone in 2007 while he was still in high school. A few years later, he made a similar exploit by hacking the PlayStation and releasing his code online, which resulted in a lawsuit from Sony.
Since then, Hotz had stints at several high-profile tech companies such as Google, Facebook, SpaceX and most recently at the AI firm Vicarious at which he worked as a researcher until last July according to his LinkedIn profile.
Now we learn that Hotz was approached by Tesla to help develop the company’s Autopilot system, according to a video released today by Bloomberg (see below).
Hotz was apparently not a fan of Mobileye’s system and thought he could build a better one by himself. According to Bloomberg, Musk wanted to contract Hotz to develop the system for Tesla or to simply work for the company:
“Frankly, I think you should just work at Tesla. I’m happy to work out a multimillion-dollar bonus with a longer time horizon that pays out as soon as we discontinue Mobileye.”
“I appreciate the offer. but like I’ve said, I’m not looking for a job. I’ll ping you when I crush Mobileye.”
Musk answered with his well-known one-word email “OK”.
The now 26-year old hacker decided to found his own company, comma.ai, to develop level 3 autonomous driving systems, which translate into about 99% self-driving.
He installed the first version of its system on a 2016 Acura ILX. He equipped the vehicle with an off the shelf Dell 21.5 inch 1080P display, which he points out is bigger than Tesla’s 17 inch display in the Model S but lacks capacitive touch.
Although Hotz’s system currently relies on a LIDAR sensor, he says that the commercial version of the product will use 6 smartphone cameras costing approximately $13 each. This approach is in-line with Tesla’s since Musk recently said that LIDAR sensors are not necessary for fully autonomous vehicles.
Tesla current hardware suite consists of a forward-looking camera, a radar, and 360 degree sonar sensors. Its second generation Autopilot hardware suite from Mobileye is rumored to rely more heavily on cameras.
Hotz aims at selling its self-driving package for about $1,000. Tesla equipped each of its vehicle with the Autopilot hardware, but it sells the “Autopilot Convenience Features” for $2,500.
On the software side, it’s not entirely clear how Hotz’s system differ from Tesla’s, but it appears to rely more heavily on machine learning. Tesla’s system is known to be self-improving, but Hotz’s comma.ai system aims at driving like a human as much as possible. For example, after the car had monitored 10 hours of driving, he says it will drive like a person who has been driving for only 10 hours.
Update: Musk and Tesla responded to the claims made by Hotz and Bloomberg – Tesla corrects claims made by Geohot/Bloomberg about the Autopilot and Mobileye
Update 2: Following this incident, we published several reports on Tesla hiring top SoC talent for the Autopilot program including our exclusive report that Tesla hired legendary chip architect Jim Keller from AMD as new “Vice-President of Autopilot Hardware Engineering”, andPeter Bannon followed him at Tesla just a few weeks later and finally our report on Tesla poaching a team of high-profile chip architects and executives from AMD to develop the next generation Autopilot under Keller’s leadership.
Here’s the video of Hotz showing his car to Ashley Vance, a Bloomberg journalist and recent author of Musk’s biography:
Featured image: Bloomberg by Peter Bohler
Subscribe to Electrek on YouTube for exclusive videos and subscribe to the podcast.