Posted on June 21, 2018 by Charles Morris
Tesla’s Autopilot system has been through several versions, and will surely go through several more. While the company can perform software upgrades remotely, it can’t (or at any rate, doesn’t) retrofit hardware components of Autopilot to older cars. The result of this is that there are vehicles with three different iterations of Autopilot on the roads.
Above: Tesla's Model S (Image: Tesla)
Lex Fridman, a research scientist and artificial intelligence expert at MIT, is a co-author of the MIT Autonomous Vehicle Technology Study (MIT-AVT), which is collecting real-world driving data and using AI to analyze the data and gain insight about human-AI interaction in driving.
Above: A look at MIT's Autonomous Vehicle Technology Study to date (Source: MIT Human-Centered AI)
As part of his work, Mr. Fridman has prepared a set of nifty charts, based on Tesla’s quarterly investor letters and other data, that illustrate how many Tesla vehicles have been delivered, and how many of them have each version of Autopilot hardware. There are now over 269,000 Autopilot-enabled Teslas on the roads, and more of them have hardware version 2 than version 1.
Above: A look at total Tesla vehicle deliveries and Autopilot hardware from Q2 2008 - Q1 2018 (Source: MIT Human-Centered AI)
Timeline of Autopilot milestones:
Fridman has projected the delivery figures forward to April 1, 2020 [significant Tesla milestones often seem to fall on April 1 for some reason]. The chart shows two alternate future scenarios: Projection A assumes average production of 3,500 Model 3 per week; Projection B assumes average production of 5,000 Model 3 per week. Both assume weekly production of 1,697 Model S and X, a prediction based on the levels achieved in Q1 2018.
Above: A historical look at Tesla's vehicle deliveries including a forward-looking forecast through 2020 contingent upon Tesla's production levels (Source: MIT Human-Centered AI)
The more optimistic Projection B has Tesla delivering a million vehicles by March of 2020 - an impressive achievement in anyone’s book. However, Fridman’s main interest is the present - the 269,000 Autopilot-enabled Teslas now on the road represent a powerful opportunity to study how AI can help save lives through human-robot interaction, and that’s the goal of the MIT-AVT study.
Above: Fridman gives some background on MIT's Autonomous Vehicle Technology Study (Youtube: Lex Fridman)
Fridman and his colleagues invite all Tesla owners to help with the research by taking (and sharing) his team’s detailed survey on Autopilot.