Fully autonomous vehicles are doomed — here’s why
Originally published on the Tobias & Tobias blog.
I’ve been thinking about the future of autonomous vehicles quite a bit lately and I’ve come to a controversial conclusion: I don’t think that we will ever see fully autonomous cars on our roads.
This may seem strange, given how many companies are working on self-driving cars. But I’m not talking about whether we will get a bit of self-driving. I’m thinking of fully autonomous vehicles — that don’t need a steering wheel — but that can perform the full range of duties that we expect from our cars today.
A huge amount of time, money, technology and brainpower has been poured into the development of autonomy so far and the results remain far from fully satisfying. Many current systems are clumsy and fail to work in poor conditions — like heavy rain. They are also all limited to operation in specific environments, for example, highway driving, where vehicle interactions are simplified and pedestrians should be absent.
Focusing on a limited set of environments is a logical step in tackling a complex problem, but the progress in these limited cases can blind us to the extent of the challenge that remains in tackling other situations. A few examples should help to make the extent of the problem clear.
Pedestrians
First, consider how autonomous vehicles will interact with pedestrians trying to cross city streets. When crossing the road, we don’t always simply wait for a space before crossing, especially in queuing traffic. Instead we rely on non-verbal cues from drivers that they have seen us and will let us go. How will this work with fully autonomous vehicles? How will the car signal to us that we should go ahead and cross? What will stop us from simply crossing anywhere and assuming that autonomous cars will have to stop?
Tricky Roads
Second, think about the full range of road types that cars are able to drive on. One of the key benefits of autonomy is the promise that it will save thousands of lives lost in accidents. However, the bulk of these fatalities are not on good-quality highways and city streets — they occur on rural roads. These are a much more complex prospect for autonomy, with more varied surfaces; poorer road markings and signage; and more complex curves and intersections.
At the extreme end, rural roads include a host of special cases that a fully autonomous solution would have to deal with. One example is single track roads with passing places. How will autonomous vehicles tackle these? If two cars meet, which one will reverse and how will they tell the difference between legitimate passing places and soft verges that they will get stuck on?
Special Hazards
Another unusual feature they might encounter is a ford — a shallow stream of water crossing a road. These are usually safe to drive through, but how will autonomous vehicles deal with them? Will they be happy to drive through water if it obscures the road surface? What if there has been heavy rain and the water is higher than usual — will they read the flood gauge and refuse to cross if the water is too high?
These may seem like spurious examples — rare considerations in the modern world. But if we consider the quality of roads around the world, it is clear that there are many more roads of poor quality than there are of good quality.
From these few examples, it should be clear that there is a huge gap between the initial steps towards self driving that have been demonstrated so far and achieving full autonomy everywhere, and that the work to be done is far more complex than the work already completed.
It is tempting to assume that, since we are making incremental progress, it is inevitable that we will finally reach the goal of full autonomy, but this is a mistake: Incremental progress doesn’t help us to predict future success. For example, if I set out to walk from London to Paris, I will make incremental progress for the the first fifty or so miles. That doesn’t change the fact that I will be stumped when I reach the English Channel.
Covering all of the possible cases might turn out to be technically intractable with current technology. But even if it is possible, it may not be economically feasible. It is no good being able to achieve full autonomy if we can’t afford to pay for it.
Implications
One possible outcome is that we abandon the idea of full autonomy and continue with cars as we have them today: self-driving on good roads and in good conditions, but controlled by the driver in other situations to permit flexibility. Fully autonomous ‘transport pods’ might well exist, but would be restricted to taxi-like duties on restricted roads.
However, another possible outcome exists: we redefine what it means to be a car; redefine what a road is; and redefine what people are allowed to do.
In this version of the future, people are banned from driving for their own good. ‘Transport pods’ would pick us up and take us to our destination. But they will only operate on good quality roads — we will have to walk the rest. In the process, what we think of as a car today would disappear.
Since the quality of the roads is so important, governments would be expected to spend more money to ensure that roads are better maintained. Problematic features would have to be replaced — if your community lies on a narrow road with a ford, then the road must be widened and a bridge built.
To eliminate the problems of dealing with pedestrians, their freedom would be limited: for example, only allowing people to cross at designated points. This may seem like an unlikely and authoritarian step, but it would hugely simplify the problem; make traffic flows much more efficient; and improve pedestrian safety — a powerful set of benefits that might be used to push through such a change.
No one can predict exactly what direction we will take over the next few decades, but it seems highly unlikely that we will ever achieve a fully autonomous solution with anything that looks like today’s roads and cars. Are we prepared to accept such a radical change?
