Let’s Do Variably-Autonomous Cars Instead
It seems like all the big automotive and technology companies have a self-driving car in development. Google and Tesla are often in the news but they have plenty of competition. The common goal is full autonomy: give the car your destination and it will navigate the roads, traffic, pedestrians, weather conditions, and occasional randomness get there. And it will be a tremendous achievement when our car doesn’t need us any more.
But what if variable autonomy is better for commuting and intra-city travel?
Fully autonomous cars come with a big set of challenges: collision liability, self-contained sensor packages that have to see everything, software bugs (or inadequacies), hardware failures, and the general ability to handle the randomness inherent to driving. California has drafted rules that a person must always be at the wheel, at least for the next three years. And since there are so many companies independently developing self-driving cars, each car will have a different set of design trade-offs and weaknesses. It might be a while before this initial crop of self-driving cars achieves actual autonomy due to laws, insurance, and social inertia, even if they are technically capable. Thus keeping them semi-autonomous in practice.
This is where intentionally variably-autonomous cars come in.
What if our self-driving revolution was based on swarming cars whose autonomy depended on the size of swarm they were in? That is, as the swarm grows, less input is needed from each driver to the point where one can safely read a book or use a cellphone when their swarm is large enough.
The general idea isn’t new. Volvo is working on platoons and convoys, and Nissan is already thinking along the lines of swarms. Birds, fish, and insects have evolved to appreciate the efficiency of flocks, schools, and swarms and that same efficiency is also available to us.
The swarming car I’m talking about has most of the sensors and intelligence of a fully autonomous car (taking advantage of the 80/20 rule) but would also use ad hoc wireless networking to share the sensory and computational load amongst other members of the swarm, instantly communicating the collective view to all members. In this arrangement, the swarm would have sensors on its outer edge with computation and communication on the inside: more sensor and computing capacity than any single car could ever have, growing as the swarm does.
Want to create a swarm? Drive up to a car that can communicate with yours. They’ll figure it out. Want to break off from a swarm? Hit the brakes or turn the direction you want to go. The other cars will get out of your way. Then you and any others who also broke off will form a new swarm going its own speed and/or direction.
When driving alone like we do now, the driver has full control (or whatever degree of control the local law and insurance rules permit). That wouldn’t change. Any time the driver is required to take control, such as when there’s a technical problem or the swarm becomes too small, there’d be an unmistakable notification: a problem our smart-phones have already solved. If the driver doesn’t immediately take control, the car would pull over to safety — a feature that self-driving cars already have and which could prevent many of the accidents we see now due to inattention.
In terms of the underlying technology, we’ve already done wireless adhoc networks, distributed fault-tolerant computing, sensors, AI, open protocols, intuitive UI’s, and we have small, stable, and cheap vehicles. We’re even well on our way to fully-autonomous cars. So we can do this.
Swarming variably-autonomous cars could become ideal commuters and even become an upgrade to the typical mass-transit experience, especially if the car is as small and stable as the Lit Motors C-1. At scale, a car like that would become quite cheap, perhaps with leases down to the price of a monthly transit pass ($91 to $170 per month in Vancouver), especially if its sensor package, “brains”, chassis, and power source are standardized. At the C-1's size, and with coordinated high-density swarms, we could fit ten swarm-cars in the same space that one takes up now, and more as speeds rise. That saves many billions of dollars on highway and bridge expansion and cuts commuting costs that are often more significant than we realize.
With their high-quality sensors and in-built intelligence, these cars could be safely operated at higher speeds than now, further increasing highway throughput. Combined with their coordination, and swarming cars would be able to flow around regular-old vehicles much like a school of fish splits around a predator or a flock of birds flows around a skyscraper. Backward compatibility built in.
Many people already have dedicated commuter cars: compact, fuel efficient, and relatively bland. They’re tools. Similarly, many people depend on mass transit, car-shares, taxis, and Ubers, which are also all tools. Done right, the variably-autonomous small swarming car can replace all of them and greatly simplify our lives without taking away the freedom and pleasure of driving our own car. Besides, our roadsters and SUV’s don’t need to go anywhere; they’ll remain as fun and functional as ever.
How? We can start with a variety of simulations, then transition to video games, then into driving robots, and then perhaps into closed-course competition much like how F1 racing advances trickle down into the mainstream. The process could be very quick if we focused on it: we already do simulations, video games, autonomous robots, robot wars, and drone races.
Trust is the key to adoption; many of us need to see it to believe it. And by taking the simulation path, we don’t need to wait for a well-heeled corporation to take the lead. We can take a page from the tinkerer and Open Source handbooks and do it ourselves.
