Real self-driving: when a person can fully turn their attention elsewhere and leave control of their vehicle to a computer. This breakthrough will transform our lives, unlocking billions of hours of our potential to put towards something better than traffic.
But real self-driving requires perfection — a driving model that performs better than a person, requiring no driver oversight or last-second intervention, and a software and hardware system that never fails, with a formally proven run-time and full redundancy.
There is a path to perfection, one that requires extraordinary focus. At my previous company, Pure Storage, we achieved it in our 6th year, reaching 7 9’s (i.e. 99.99999%) of uptime and 18 9’s of data reliability — the closest any enterprise storage company has gotten to perfection.
Today with Ghost, we are charting a new path to hardware and software perfection to deliver real self-driving to hundreds of millions of people. To do this, we’re focusing on highways first.
Perfection Demands Simplicity: Highways
The first step to perfecting a software system is to simplify the problem.
To date, the most prominent self-driving companies have tried to solve all the complexities of door-to-door urban and suburban driving before commercializing a product. This expansive challenge has proven to be difficult, and has yet to deliver a product safe enough for mass consumption.
Self-driving demands perfection. Perfection is unachievable everywhere at once. There are functionally infinite variables across geography, environment, pedestrian mix, weather, and more. To get real self-driving somewhere, we need to simplify the problem.
Highway driving is the simplest and most common form of driving. Two-thirds of all miles driven in the US are on highways, in an environment designed for high speeds with minimal complexity.
Focusing on solving highway driving dramatically simplifies the technical challenge of self-driving, making perfection possible in just a few years. We can deliver real self-driving, exit-to-exit on the highway, and solve a real problem for everyone driving on today’s roads.
Perfection Demands Proof
After simplifying the problem, you next must engage in a rigorous process of development, testing, and refinement to perfect a software system.
Is it safe? It’s the only question that matters in autonomous vehicles, and it is the one question that no one can answer. To date, we have lacked proof. There is no meaningful measurement of quality or safety to define a real standard. We are here to change that with a new approach.
The first step in proving safety is proving your driving model against a standard. Most people agree that if a self-driving car performs better than a human driver, it is fit to drive on today’s roads. That’s why at Ghost, humans are the standard against which we measure ourselves. We numerically compare our driving model to how real people safely navigate the real world so that we can show when we’ve achieved driving that beats the human standard. When Ghost is ready, drivers will have hard proof that allows them to confidently give Ghost control.
The second step in proving safety is proving your software execution, ensuring your software does exactly what you write, with no errors. We are the first self-driving company to formally verify its real-time driving system, a process typically reserved for aerospace and defense industries, to ensure perfect software execution. Through formal verification, we keep drivers safe from software bugs and glitches.
The Path to Perfection
Self-driving is the pursuit of perfection. It can only be achieved with dramatic simplification and continuous measurement against a real-world standard. We believe the highway is the fastest way to get there.