Update: Tesla’s Fleet Learning
How the Silicon Valley software company will win the autonomous car war
Last week, Hackerbay Research released Tesla’s competitor analysis report. We release an update today, because our proprietary innovation miner (a large-scale search engine for innovation projects) discovered a new keyword in the field of Artificial Intelligence: Fleet Learning. After extensive research, we found out that Tesla’s 200.000 terabyte per day network effect is actually a new branch of machine learning, invented by Tesla and called Fleet Learning.
Fleet Learning — network effect based machine learning
There are several strategies to get to a self-driving car and Tesla decided to focus on big data.
Tesla was able to restructure their hardware architecture after Mobileye services were abruptly terminated in 2015. The argumentation of both companies got ugly and public in September 2016 after Tesla blamed the hardware for the deadly car accident of a Model S with active autopilot. Mobileye got acquired by Intel for $15 billion in March 2017 — the largest exit in Israel startup ecosystem.
Since this event Tesla focussed on data as their core competitive edge in the self-driving car game.
Tesla is focussing on the 2nd law of Hacker Culture as competitive edge in a digital world: Metcalfe’s Law. The law was developed in the context for computer networks, but it stands as a concept applicable to systems both within and without the technological sphere– where any network of any sort can be created. Tesla embodies Metcalfe’s Law and leverages it perfectly to spur on their expansion and success with their autonomous driving technology. The more cars are connected to the cloud, the more data Tesla receives from those vehicles. Therefore, the more well-informed the service becomes with the capability to provide more information, a better service and a much higher level of reliability.
The whole Tesla fleet operates as a network. When one car learns something, they all learn it. That is beyond what other car companies are doing. — Elon Musk, Tesla (Source)
The Silicon Valley based company described the fleet learning algorithm in an official blogpost of 2016:
When the car is approaching an overhead highway road sign positioned on a rise in the road or a bridge where the road dips underneath, this often looks like a collision course. The navigation data and height accuracy of the GPS are not enough to know whether the car will pass under the object or not. By the time the car is close and the road pitch changes, it is too late to brake.
This is where fleet learning comes in handy. Initially, the vehicle fleet will take no action except to note the position of road signs, bridges and other stationary objects, mapping the world according to radar. The car computer will then silently compare when it would have braked to the driver action and upload that to the Tesla database. If several cars drive safely past a given radar object, whether Autopilot is turned on or off, then that object is added to the geocoded whitelist.
When the data shows that false braking events would be rare, the car will begin mild braking using radar, even if the camera doesn’t notice the object ahead. As the system confidence level rises, the braking force will gradually increase to full strength when it is approximately 99.99% certain of a collision. This may not always prevent a collision entirely, but the impact speed will be dramatically reduced to the point where there are unlikely to be serious injuries to the vehicle occupants.
Competition Demystified
Our research about competitive advantages is deeply inspired by New York based hedge fund legend and value investor Bruce Greenwald. In his book Competition Demystified, Greenwald talks about 4 dimensions of competition, which he analyses before investing in a stock or asset: Brand, Economics of Scale, Proprietary Technology, Network Effects.
In our update, we now could go into details with Tesla’s approach and will demonstrate how Tesla uses technology not only to gain competitive advantege in range and performance (Moore’s Law), but invests heavily in cloud computing and data analytics to win Metcalfe’s law’s as well.
May 12, 2017 Trent Eady, started a discussion about Tesla’s competitive advantages in Transmission group of Udacity’s self-driving car nanodegree on Facebook, the best community I know about self-driving cars.
Trent Eady posted a very interesting article 9th of May 2017: Tesla Leapfrogs Self-Driving Competitors With Radar That’s Better Than Lidar The article is the foundation for the discussion, if Tesla will win against all other competitors and therefore be the most valuable car company. Trent built 18 thesis about doing so and I will highlight some:
(2.) The opportunity will be even larger for Tesla if it achieves full self driving in its production cars long before competitors. For one, it will create essentially unlimited demand for Tesla’s cars, including the high-end Model S and Model X. For another, Tesla could charge higher fares on the Tesla Network, since it would be competing with human drivers.
(16.) Incumbent car companies can’t immediately solve their software shortcomings just by acquiring or partnering with software companies. No software company besides Tesla has experience in fleet learning at production scale or provides over-the-air updates for driver assistance software. Tesla is the only company with experience in this area in a production environment. It will take time for any competitor to develop this competence.
(17.) Combining competitors’ reliance on lidar, car manufacturer’s stated production timelines, their slow development cycles, and competitors’ lack of automotive software expertise, it seems likely that Tesla will achieve full self-driving in its production cars one to two years before any competitor.
Source: https://seekingalpha.com/article/4071192-tesla-leapfrogs-self-driving-competitors-radar-better-lidar
Tesla’s Core KPI: 6 billion cumulative autopilot miles to successfully complete the beta test
Elon Musk wrote in his Master Plan, Part Deux (July 2016) that the worldwide regulatory approval will require about six billion autonomous miles to improve the fleet learning algorithms.
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 true self-driving is approved by regulators. We expect that worldwide regulatory approval will require something on the order of 6 billion miles (10 billion km). Current fleet learning is happening at just over 3 million miles (5 million km) per day.
Moreover, Elon Musk publicly explained what beta means to a software company like Tesla:
Innovation Mining: Tesla’s Autonomous Miles
Tesla is not disclosing autonomous miles to the public on a regular basis, but Hackerbay’s innovation miner, did find some public data about cumulative autopilot miles by Tesla:
April 2016 (Source Tesla, Twitter):
October 2016 (Source Elon Musk, Twitter)
December 2016 (Source Tesla Third Quarter 2016 Update)
Tesla vehicles have already been driven over 3 billion miles,including more than 1.3 billion miles logged by vehicles with Autopilot hardware.
The data in December 2016 is not autonomous miles, but it seems like logged miles in general on hardware. So, the data is valuable, but not similar to cumulative Tesla Autopilot miles.
According to the Tesla Master Plan, Part Deux that refers to 3 million miles of fleet learning per day (which would result to about 400–500 million miles of cumulative autopilot Data in December) is seems that fleet learning is not the same miles like the 1.3 billion miles reported in the shareholder update.
Forecast Innovation Mining: 1 billion miles in June 2017
If the 3 million miles of fleet learning per day since October 2016 stayed, Tesla will hit the 1 billion miles of fleet learning probably in June 2017. Here’s my predictive calculation open-sourced: Google Sheets Link
Conclusion: Tesla Model III will accelerate autonomous miles
Tesla is in a strong competitive position, because with the release of Model III and full autonomous hardware, the car company will probably hit the 6 billion miles way faster than expected (2021).
Tesla’s speed and culture is a good thing for the market, because self-driving cars have the potential to make transportation safer, cheaper and more reliable.
