Tesla owners have already driven 1 billion miles with Autopilot engaged and if Tesla sustains its current vehicle delivery rate, Autopilot miles will reach a massive 2.3 billion by the end of 2019.
Lex Fridman, a research scientist at MIT working on human-centered artificial intelligence explains in detail how their team arrived at the 2.3 billion miles estimate.
I started with the number of Tesla vehicles delivered by quarter and organized by Autopilot hardware version.
Next, I did an estimate of per-day deliveries dating back to 2008 in a way that fits the quarterly reported delivery numbers.
Finally, I enumerated the number of miles driven in each vehicle under manual and Autopilot control.
“Today it’s over 1 billion miles. By end of next year, it’ll be over 2.3 billion. All of us working in autonomous vehicle research want nothing more than to save lives. Happy holidays & good luck,” Lex Fridman tweeted yesterday.
According to Fridman, Tesla will have 1.48 million vehicles on the road by the end of 2020. Tesla already has more than 440,000 vehicles on the road and the company manufactured more than 80,000 vehicles in the third quarter of 2018.
If Tesla’s production averages more than 8,000 units/week over the next two years, the electric car maker will easily reach Fridman’s estimate of 1.4 million vehicles on the road by the end of 2020.
The fleet number is conservative because Tesla is planning to start local production in China by the end of 2019.
With Tesla adding more than 6,000 vehicles to its fleet every week,
Tesla CEO Elon Musk said that automakers will need 6 billion miles of data to get regulatory approval for self-driving.
“As the technology matures, all Tesla vehicles will have the hardware necessary to be fully self-driving with fail-operational capability, meaning that any given system in the car could break and your car will still drive itself safely. It is important to emphasize that refinement and validation of the software will take much longer than putting in place the cameras, radar, sonar and computing hardware.”
“Even once the software is highly refined and far better than the average human driver, there will still be a significant time gap, varying widely by jurisdiction, before
Elon Musk gave us just an estimate and the real requirement is anybody’s guess at the moment.
Auto companies working on self-driving technology must provide hard evidence for them to have any chance of convincing drivers all over the world and to obtain necessary regulatory clearances.
But when it comes to data, Tesla is
Todd Shaver, Founder
“It’s going to take many years before enough people reach the right point on their comfort zone and vehicle trade-in cycle to make the autonomous world the status quo. And in that intervening time, whoever has the best track record will eventually win.”
He added, “Since self-driving technology improves as more miles are driven, Tesla’s 1.5 billion miles of road tests are already ahead of everyone else. By the time drivers start making meaningful comparisons, this system will be the gold standard.”
Tesla has a triple advantage in the self-driving car race.
2.Ability to update software over the air
3.Real world Autopilot miles, not simulation
By installing the hardware required for self-driving in its cars, Tesla has turned the self-driving challenge into a software problem.
With over the air-updates, Tesla can continuously improve its software and every day Tesla is accumulating real-world data on Autopilot performance.
In October this year, Elon Musk announced that Tesla’s custom AI Chip that offers 500% to 2000% performance improvement will be ready in six months.
If data is what is going to tip the regulatory scale and convince drivers, then it will be Tesla that has the edge in the self-driving race. Tesla’s we don’t want to use Lidar approach may spell trouble for the company, but fortunately, Tesla has all the data it needs to know if it has taken the wrong approach.
About the MIT-AVT Study
The MIT Autonomous Vehicle Technology (MIT-AVT) study was launched “to understand, through large-scale real-world driving data collection and large-scale deep learning based parsing of that data, how human-AI interaction in driving can be safe and enjoyable.”
You can read more about the study by visiting their blog