Autonomous Cars Could Save More Than 30,000 Lives

A few weeks ago, I hailed a car on an app on my phone.  A Jaguar soon pulled up about 20 feet from me, and the app notified me that I had five minutes to meet the car.  I tapped on the unlock button in my app and the door handle popped out.  I entered the car and was greeted by a voice welcoming me to the vehicle and explaining some basics.  I belted myself in and then clicked the “go” button on the center console.  The car pulled out and started driving me to my destination.

There was no one in the driver’s seat.

This was an autonomous car from Waymo.

The drive was comfortable and remarkably dull.  The car got me to my destination and drove better than most of my friends and family members.    

The novelty of the car driving itself wore off quickly.  I was soon pecking away at my phone.  I did enjoy using Google assistant to play music of my choice. I chose “Country Roads” by John Denver and may or may not have started singing loudly in the car.

Robotaxis are already here.   They have shown that they can drive safely in certain contexts. The share of miles driven by computers, relative to humans, is rapidly increasing.  Tesla, General Motors, Comma.AI, and other auto manufacturers are rolling out increasingly capable autonomous systems to assist or replace humans in driving. 

Autonomous Cars Should Be a Public Safety Priority

I am incredibly excited by the positive potential of autonomous driving systems. 

Increased capabilities of autonomous driving systems could potentially drive a massive decline in automotive fatalities. Currently, car accidents are one of the top killers [1] of people in the US, tragically ending the lives of ~44,000 in 2023[2].  And Waymo reports that their autonomous vehicles are involved in injury causing accidents at 15% the rate of human drivers[3].   If we extrapolate this safety figure:

• Self-driving cars could end up saving ~37,000 lives per year in the US. 

• That’s almost as large as eliminating deaths from breast cancer (breast cancer claims ~42,000 people per year)[4].

• Globally, ~1.2 million people die in automobile crashes every year.  It’s possible that autonomous driving could save more than 1 million lives per year. 

• I do not believe that an 85% improvement in driving fatalities is a ceiling on autonomous vehicles.  As the proportion of miles driven shifts to safer (autonomous) drivers, there may be an exponential decline in accidents because there are fewer bad (tired, inebriated, bored, preoccupied, etc.) drivers on the road. 

There are a few downstream ripple effects: 

• Overall health care costs could decrease with fewer accidents.

• Car insurance costs may decline.  Swiss RE has adopted[5] this view. 

We do not know how quickly this transition can take place.  Different companies have different approaches to self-driving, and potentially different safety profiles, and most assuredly different economic models.

As we understand it, currently there are two competing technological approaches to solving autonomous driving. 

Approach #1: Geo-Fenced, Precision Mapped Areas + Vehicles with Expensive Sensors

This is the apparent approach that companies like Waymo and GM Cruise have taken.  They choose a particular area, map it with a high degree of accuracy, and then train their vehicles, which have an expensive array of sensors (including LIDAR, cameras and radar), to drive with very high quality in that geo-fenced area. 

Waymo is now scaling operations to new cities.  GM Cruise, however, has been involved in a fatal accident which has caused regulators to suspend[6] their license in San Francisco for some time.

The potential negative of this approach is its high upfront expense.  We estimate that Waymo spends ~$100-200K per vehicle deployed.  At that level of capital expenditures, and associated maintenance costs, it could take ~10,000 rides (at current prices) to make an economic profit on the investment.  I’d expect Waymo, and others, to be highly focused on a much cheaper version for future generations to improve their path to profitability.    

Approach #2: Vehicles with Cameras + Good Enough Maps

Tesla and Comma.AI are two companies that focused exclusively on cameras as the necessary sensors for an autonomous vehicle.  They believe that since humans can effectively drive with two cameras (their eyes) on a swivel, it should be possible for autonomous vehicles to do the same without the more expensive Lidar technology.  Tesla’s technology is embedded in the cars they sell, while Comma.AI sells a $1250 accessory that can be added to most existing cars, and takes over steering, braking and acceleration of a car. 

Both companies are making progress on their efforts. The software has significantly improved over the past year.  Real world data related to the autonomous driving systems are the key inputs that allow the companies to improve the driving models, and both companies send data from their “fleets” back to corporate headquarters to help train and refine the autonomous driving systems.  Morgan Stanley’s Adam Jonas estimates that worldwide, Tesla could be driving around 100 billion miles per year by the end of 2025, which is a lot of data.

The challenges for Tesla and Comma.AI is more difficult than those that companies using the first approach have, given the latter’s sensor heavy approach.  Tesla and Comma.AI have less precise sensor and map data, and they are operating in many more geographies with far more varied conditions (snow, rain, fog, round abouts, terrible highways and exits, etc.). 

The camera approach is also much more scalable.   If the software improves past a certain point, there are already millions of Teslas on the road that could become autonomous, and tens of millions of cars that could add Comma.AI’s accessory to become autonomous.

What Does a World with Millions of Autonomous Cars Look Like?

There should be significantly less traffic injuries and deaths. Though, we will probably be far less forgiving of accidents that happen because of autonomous mistakes.  The autonomous vehicles will have a much higher bar of safety expectations, with a practically perfect record, to survive the onslaught of negative publicity that may occur when they are involved in a fatal accident.  Even if driver deaths go from 41,000 per year, to 1,000 per year because of improvements in safety from autonomous vehicles, I’d expect there to be articles and politicians bemoaning “the 1000 people killed by greedy companies, and their death cars.”

I expect Robotaxis will be ubiquitous in every major city in the near future.  Uber and Lyft operate two-sided marketplaces that connect drivers and riders.  Right now, they benefit from virtuous network effects, which leads to shorter wait times resulting in more riders and drivers etc.  In the future, those companies, and others, may instead be connecting autonomous cars and riders. It’s unclear how the economics in that marketplace will change at that time.

We expect non-autonomous car sales to plummet.  Why own a car that sits idle for most of the time, when there are fleets of safe, driverless cars, willing to take you to your destination?  If you want your own car, why wouldn’t you buy an autonomous car that can produce income when you aren’t using it (like Airbnb).  Automakers who do not develop their own self driving software, will likely have to license software/hardware from those that have it. 

[1] Source: Road Traffic Injuries and Deaths—A Global Problem (January 2023)

[2] Source: Traffic Crash Deaths In 2023 Remain At ‘Devastatingly High Level’ (February 2024)

[3] Source: Waymo (December 2023)

[4] Source: Breast Cancer Statistics (February 2023)

[5] Source: Comparative Safety Performance of Autonomous- and Human Drivers: A Real-World Case Study of the Waymo One Service

[6] Source: GM’s Cruise Loses Its Self-Driving License in San Francisco After a Robotaxi Dragged a Person (October 2023)