Is autonomous driving chasing the wrong dream?
Back in 2019 Elon Musk predicted that by 2020 that Telsa’s autonomous driving would be full baked to the level that robot taxis would be cruising around. The problem is that driving in cities is hard, there are lots and lots of edge cases. The real goal here for these autonomous driving elements in city is the, very lucrative, shift towards fractional ownership/rental, imagine Uber/Lyft but for every local journey. This means that the cars must be fully autonomous, operating on their own as a single element, just like drivers do, because each car is travelling to a different destination that just happens to overlap for a fraction of time.
Driving in cities is hard, from other road users, pedestrians, cyclists, flying litter and in the case of Paris, the insanity of Etoile. There are just too many edge cases that happen every single day, and there are accidents, lots of accidents, driving aids can help, but maybe this challenge is beyond what AI can do in the next decade. It is why I’ve advocated for regulators to create headless Metaverses before such things are allowed on the roads, and there are folks out there creating traffic simulations which could help.
So what about another dream? Looking at a problem which has less variability, where cars and lorries could collaborate, where accidents are often caused by tired and distracted drivers due to long periods without significant change and one where speed variability is a larger impact on CO2 consumption/energy efficiency than the speed itself. I’m talking about freeways, motorways, roads where there is a divider between traffic that is going in different directions, which has clear on an off ramps which aren’t at right angles. Roads designed for distance and moving large numbers of cars from A to B. If we accept that city based, and rural, driving will require augmented rather than artificial intelligence, can we take a new approach to long distance driving that can truly deliver?
If we look at this problem as a collaborative problem as opposed to each vehicle being an autonomous entity we can quickly see the potential. This isn’t a new idea, I was presenting at a Mobile Data Management in 2004 and several of the papers are on this topic, and since then many people have proposed collaborative vehicle networks as a way forwards. A challenge is that such efforts really require all vehicles to be involved, otherwise you still have edge cases. But given alternatives like digging loads of tunnels its certainly something to consider.
In a collaborative situation where vehicles join a network, each able to communicate and provide information you can address the challenge of safety and efficiency in very different ways than a lone vehicle could manage. Take the Accordion Effect where an event causes traffic to slow down, which causes other vehicles to break harder, and then when you get to the point of the road where you are travelling the slowest you look around and realize: there is nothing really causing this. A collaborative ecosystem for the road network would be able to identify the initial cause (or help prevent it) and then be able to smooth out traffic behind so the maximum average speed is maintained. Reducing delays, reducing breaking, and thus reducing energy consumption.
In a collaborative ecosystem we could also institute something like TCAS (TL;DR: software on an airplane that stops it bumping into other airplanes), so if a car is forced to break hard and fast, for instance due to an animal, flat tyre or alien invasion, then that signal can be rapidly propagated to all surrounding vehicles to negotiate out the most efficient solution, informing the car straight behind that it also needs to break. This would save significant time over even existing automatic breaking solutions which can only work after the breaking has had an impact, not at the moment of that impact.
Collaborative Autonomous Driving alters the dynamic, because like any good Collaborative Data Ecosystem it enables the individual AI within a car to not just see what it can see, but to see the context within which it sits and the context of the full journey. Instead of the AI trying to optimize in competition with other AIs when there are issues it can be done collaboratively, so for instance cars quickly creating the 2D map of where each of them will turn to avoid a collision, rather than one AI choosing “left” and the one to it’s side choosing “right”, both thinking there will be a space, but resulting in an avoidable collision. It also can help in cases where there was reduced visibility and prevent the sort of pileups that happen, and which in poor conditions a lone AI will struggle to address, particularly if just using vision not radar.
The problem with this approach is that it isn’t ‘sexy’, it won’t result in fleets of robotaxis that don’t require drivers. What it will enable is more efficient and reliable distribution of goods via trucks, safer and faster long distance travel, and significantly impact CO2 emissions and energy consumption generally. It would require a focus on standards, and need multiple car companies to come together, and a committed government to trial and then scale such an approach. It would require insurance companies to value the safety of autonomous long distance driving and thus reduce premiums for when it is used.
I’m personally not convinced that current AI approaches will be able to handle driving around Etoile a rush-hour, or the streets of Wolverhampton late on a Friday night. But I do think that collaborative systems for transportation within the semi-closed environment of a motorway or interstate are achievable if there is the willingness to drive collaboration. There are other advantages for collaboration if done at scale, with the ability to create “road-trains” which act as joint systems, reducing overall drag, and thus energy consumption. This sort of approach would also significantly help commuting times. Enabling people to be much more confident that they’ll get there at the right time, and as we transition to EVs enable more confidence that you’ll get there with enough charge to make it home.
Collaborative Driving Ecosystems aren’t as “cool” as full autonomous driving, but then given that full autonomous driving around Etoile is probably still a long way off, is it time to start planning
