Takeaway: Through use of redundant camera data, low maintenance fleet, custom neural network chip, and a relentless improvement culture, Tesla is poised to dominate the autonomous driving industry.
Autonomous driving is the next tech frontier, and is poised to unlock trillions of dollars of value and be one of the largest societal impacts since the iPhone. This is why Microsoft is investing into Cruise, Google has their own self driving division and even Apple is trying to join the fray.
Due to the nature of the technology, we believe that autonomous driving is set up to be a winner take all market, much like how Google dominates search today. In this blog we will review the key pillars required to win autonomous driving and why Tesla is poised for the win.
1. Redundant Camera Data
To win autonomous driving, the car must drive better than a human.
Redundant Camera Data means having both 360 degree camera coverage and plenty of data. With these combined there is a path to greatly exceed a human person’s ability to drive.
The majority of autonomous vehicle companies claim that Redundant Camera Data is not sufficient to achieve full autonomy and tout that LiDAR is on the critical path. However, several papers, one from 2018 before Tesla’s Autonomy Day presentation and one from 2020, outline that current deep learning technology can recreate the depth data that LiDAR provides, rendering LiDAR largely obsolete. LiDAR is not on the critical path for autonomy.
With simply 360 degree coverage, and the right neural network training, a car can achieve superhuman driving ability. Let’s take a look at a couple approaches.
No Redundant Cameras
GM adds cameras to their Cadillac line for Super Cruise, but it is only forward facing. Humans can look in front, to the side, and behind the car with mirrors, so this solution will never become fully autonomous.
Redundant Cameras But No Data
Google’s Waymo has cameras everywhere, but has no scale. Each Waymo has multiple $7,500 LIDARs that make the vehicle cost nearly $200,000 at the time of this writing. This makes for good demo rides, but without massive scale, there isn’t enough data fed into the machine learning system to interpret the world and drive safer than a human.
Redundant Cameras at Scale
With 360 degree camera coverage at scale, there is enough data fed into the system to unlock the latest in machine learning technology, including the ability to do trajectory planning across multiple frames (4D data). Tesla has managed to get people to pay them to give Tesla data since every Tesla since 2019 feeds them the data to unlock autonomy. This is in stark contrast to having to pay test drivers to capture a much smaller dataset, which forces them to use sensors like LiDAR as a crutch.
2. Low Maintenance Electric Fleet
To win autonomous driving, there needs to be a lasting cost advantage.
A gas car could be the first autonomous car, but the advantage would be short lived due to the cost per mile. Say a gas vehicle takes at least $1/mile to operate since gas vehicles have higher maintenance costs, higher fuel costs and a human needs to fill up the gas for basic safety reasons. An electric vehicle could cost $0.25/mile to operate due to dramatically lower maintenance costs and the ability to charge through induction (wirelessly), without paying a person to fuel up for safety reasons. At full autonomy, the electric autonomous car would completely price out the gas one and would make the gas autonomous car company go bankrupt.
Why Tesla Will Win This Core Pillar
Per Mile Efficiency
Tesla has the most efficient and lowest cost per mile EV fleet out there and is set to continue its leadership in this area. As of January 2021, there is no production EV that has beaten the 2012 Model S’s 265 miles of range and Tesla hasn’t stayed still. The current Model S exceeds 400 miles of range.
Tesla’s range obsession would not be impressive if they simply increased the size of their battery, but they have managed to increase range through better manufacturing, and the lowest drag coefficients in the industry. The Tesla Model 3 has a drag coefficient of .23, which is even lower than the Toyota Prius. This results in more miles per kwh, which ultimately reduces costs in an autonomous car.
On top of that, Tesla’s cars are designed to last a million miles, including the battery. If Waymo’s gas cars only last 200k miles, but Tesla’s cars are designed to last a million miles, the cost of the car amortized over the miles driven will be much less for the Tesla. There are no other electric car manufacturers who focused on the longevity of their batteries like Tesla has.
Car Manufacturing Efficiency
Tesla’s relentless improvement culture allows them to think of innovative solutions to save costs. Tesla managed to reduce 70 separate pieces into a single piece of casted aluminum.
They needed a type of aluminum that was strong enough to be casted into a single piece, but not brittle. This didn’t exist, so Tesla developed their own aluminum to achieve this. Without Space X’s metallurgy expertise, it is unlikely that Tesla would have the know-how to forge their own metal and come up with a solution no one else has thought of before. Efficiencies like this reduce vehicle cost and expand Tesla’s moat as the leader in having a Low Maintenance EV fleet to secure their autonomy advantage.
3. Custom Neural Network Chip
Currently $200,000 Waymo vehicles are giving autonomous rides to people in Arizona. The trunks of these cars are filled with computer servers that eat up massive amounts of energy. This means that the trips these cars provide are limited to 30 minutes, and are not able to be ‘go anywhere’ robocars.
Tesla is Winning in This Core Autonomy Pillar for Two Reasons
Speed of Chip Development
Tesla’s chips have been analyzed as being 6 years ahead of Toyota and VW, and at least 3 years ahead of the most advanced Nvidia Orin chip.
By designing their own chip, Tesla was able to leapfrog other automakers and capture a lead in hardware. This is critical to winning autonomy because whoever is first to having the most autonomous capable cars on the road will be able to win the most critical factor, which is being able to collect enough varied data to train the machine learning models to superhuman capabilities.
Tesla’s chip also has a wide energy efficiency advantage over the trunk full of servers Waymo is forced to use. Without energy efficiency it would be impossible to have a low maintenance electric fleet and automakers would be stuck with the problem of having to charge the autonomous car for 5 hours after every 1 hour trip.
Nvidia’s future Orin chip will be much more energy efficient than Nvidia’s current Xavier chips, but Tesla has not stayed still. They plan to have a 5nm chip by the time Orin is released, which will be much more powerful and energy efficient than Tesla’s current chip.
4. Relentless Improvement Culture
Without the right disruptive culture, there is no way to maintain a dominant lead in autonomous cars.
For example, a company may come out with the first autonomous car but it is a gas powered car. Then the first one would be disrupted by the electric car.
Say someone comes out with the first autonomous electric car. Another company can disrupt that company if they are able to reduce costs more than the first one.
Since autonomous driving is a subscription model, cost leadership, electric vehicle efficiency and safety are paramount to maintain the lead, and the company that has the right culture is set to maintain that lead.
Tesla proves this in spades. They continue to increase car efficiency with innovative solutions like taking the excess heat from the onboard supercomputing chip and using it to heat the car. Or casting the entire back of the car as one piece instead of assembling 70 pieces individually. Each of these improvements secure a cost per mile advantage, which ultimately results in the most affordable autonomous driving service, and Tesla’s improvement culture unlocks this.
We are living in the midst of some of the biggest technological transitions in decades. This presents enormous opportunities for investors who see it and makes it even more important to concentrate on the likely winners in this space. A first principles research approach helps us see which company is close to actually achieving full autonomous driving, and cut through all the noise and hype that the competitors are desperately trying to put out there.