Coming Clean with Automated Driving and Autonomous Vehicles
Throughout my professional life I have been trying to contribute to a more sustainable society albeit not in a direct way but in a more convoluted manner as an engineer: e.g. Smart control for energy efficient buildings that aimed at minimising the energy expenditure for commercial buildings through a more holistic control approach; Active Magnetic Bearings for a potential application in energy storage using rotors spinning at large rotational speeds; Holographic Radar, a 4D staring radar to unlock the construction of windfarms that airports would otherwise object to; and my last venture at Five.ai which started with the mission “Shared mobility for the urban area that doesn’t cost the Earth” and the ambition of developing as well as operating autonomous vehicles aka robotaxis.
Initiation to Autonomous Vehicles
Wholeheartedly I bought into the Five.ai mission of shared mobility when I joined and I still do. I was prepared to contribute in the best way I can, because, in fact, I do not like cars (even though I am moderately attracted to the beauty of and engineering in some sports cars). Two aspects motivated me: first, the potential to considerably reduce the number of cars on the road (and by extension cars in existence or even miles driven), and second, the technical challenges that tickled my inquisitive mind. The driving task is so easy to learn for humans; but it was completely unimaginable to me how a machine could do that. By the time humans are taking up driving, they would have learned to negotiate the world in all its complexity. Our “natural neural network” aka brain would have been already pre-trained for years to perform complex tasks as a foundation; adding the driving task is a mere extension to this capability.
The initial approach by most players in the space was to carve up the driving task in a classic technical manner: nested modularity, each module with tests and interfaces, most built from first principles. The resulting architecture is shown below: sensor inputs, perception (like humans are thought to perceive the world), prediction and planning, followed by low level control, i.e. steering and pedal. Nowadays, however, most companies replaced the inside of the perception and planning with an Artificial Intelligence (AI) component (called “model”) trained on large quantities of data; some companies go all the way (like Wayve): realise perception & planning as one huge model to be trained like a human, leveraging foundation models. It is obvious to see for all that autonomous or automated driving is quite fascinating for an engineer. Personally, I could and can contribute in areas such as system modelling, vehicle control or multi-object tracking given my previous experience.
Motivation for Working on Autonomous Vehicles
Engineers like myself can get quite excited by purely technical aspects of a problem, viewing a certain situation within a narrow scope and deriving pleasure from solving intricate problems with dedication to minute details, hence allowing to gain mastery in that domain¹. For most engineers there still has to be an element of purpose, though, the knowing why. To me, there has to be a chain of purpose that links my work over a chain of intermediate goals to an ultimate aim I can believe in and that aligns with my personal world view. For instance, I might work on some aspects of a piece of code for a vehicle controller that is used in the autonomous vehicle stack which in turn is used for an autonomous shuttle as part of a fleet to make personally owned cars obsolete, and by extension significantly reduces the amount of resources needed hence making our society more sustainable. For some this chain of purpose might be very different¹, e.g. leading to fulfilment by solving technical challenge(s), or making the best product in the relevant industry, or simply for commercial reasons by serving customer demands.
For me, this chain of purpose has to lead to something meaningful in a societal or existential context. We are back to “shared urban transport that does not cost the Earth”, i.e. the goal for autonomous driving is the potential of a reduction of cars on the road. Why? Simple, the global fleet of vehicles comprises about 1.5 billion cars travelling about 1 trillion miles a year with about 90 million cars being produced every year. The amount of resources consumed in producing and operating this fleet is not sustainable. Even the electrification of cars and trucks does not solve the problem²; it alleviates some aspects such as energy consumption as an electric drive train is about 3 times more efficient than an internal combustion engine; the amount of resources needed is still unsustainable.
Autonomous Vehicle, Quo vadis?
When the concept of autonomous vehicles became popular in the mid 2010s, the goal for most players was to provide alternatives to individual car ownership. I do remember cycling to work at Five.ai through the narrow streets of Cambridge with traffic in both directions and cars parked on both sides, too, leaving only a narrow lane for all traffic participants. What motivated me was that every morning I thought: if we succeed, all these parked cars will be gone and autonomous, shared taxis will chauffeur people around; hence the purpose was clear. But let’s revisit that idea with a thought experiment: everything else being equal, children to school, people going to work, shopping, to the cinema or on holiday. What would autonomous vehicles offer, what would be their purpose?
- Case A: keep the premise of personally owned vehicles but equip them with automated driving assist system (ADAS) features. This seems to be the new direction in the automotive industry as it turned out that consumers won’t let go the notion (or illusion) of individual freedom through individual car ownership; customers value features and convenience³. The purpose is to meet customer needs (or wants), i.e. a pure commercial proposition⁴. On the upside, there is some added safety and the development of autonomous driving features could potentially be a stepping stone for another aim, robotaxis.
- Case B: most individually owned cars are replaced with robotaxis. In the worst case, however, twice the amount of driving is then needed (to account for the robotaxi getting to customers and back), but there is huge potential of logistics optimisation, energy and resource reduction. The number of miles driven per person may the same but on the upside fewer vehicles are needed in total, perhaps an order of magnitude: a fleet of 100-200 million robotaxis could meet personal mobility needs, i.e. purpose of autonomous vehicles is that much less resource & energy is needed, an economic and ecological goal.
- Case C: most individually owned cars are replaced by robotaxis as autonomous shuttles carrying say a dozen people. Even fewer but slightly larger vehicles are required with the added minor inconvenience that a vehicle needs to be shared with others and it might take a bit longer to get to a destination. A much larger net gain in transport efficiency and resource reduction could be expected, again providing a purpose as an ecological and socio-political goal. Yet the number of miles per person still the same, if not more (Jevons paradox).
Simpler than Autonomous Vehicles
During the past few years working on autonomous vehicle technology, from stack to tools, I kept reading and writing more about environmental issues (planetary boundaries, etc) which altered my view a bit: a world with robotaxis would indeed consume fewer resources, but the pursuit of this vision might distract from the urgency of changes needed to maintain life on this planet; fully automated personal mobility may not arrive in time. Many alarm bells are ringing: ocean acidification, soil degradation, deforestation, desertification, to name but a few issues to become severe within a decade or two. While automated driving is challenging, there is not enough time, there are more pressing issues: energy and food security while protecting the biosphere we all depend on. If, as a society, we want to maintain a certain level of personal mobility, perhaps we need to think about investment in simpler solutions with better returns:
- For medium to long distance journeys trains are very efficient; don’t worry about range anxiety of electric vehicles and don’t make cars heavier to accommodate bigger batteries³. Electric trains are very good at distances, offer comfort and speed. You only need to get to and from the train station where even normal taxis can be adequate; ideally consumers would like to stay in their own pod all along:
- Urban planning⁵ to create 15-minute cities where it is possible to get anywhere within 15 minutes by walking, cycling and shared means of transport, or a combination of those. This can include autonomous shuttles or robotaxis, but not as the main mode of transport. Cities such as Paris under the leadership of Anne Hidalgo are making good progress in evolving away from depending on cars towards Carlos Moreno’s 15 minute neighbourhoods.
In conclusion, developing autonomous driving capability is interesting, but perhaps not the most efficient way to spend energy, resources, money and effort in order to realise efficient personal mobility⁶. For instance, the Lab of Thought under the leadership of Marco te Broemmelstroet offers many ideas on human-centred urban planning. In the meanwhile, as an individual, one should always go through an internal state machine on personal mobility: Can I avoid it, can I walk or cycle, can I take a bus or train, or hire a taxi or shared vehicle for the day and use case? If none applies perhaps use the smallest car possible with the smallest footprint, but please no SUV, the material expression of vanity and human hubris.
Footnotes
- Most people are able to compartmentalise: work on something in isolation from a bigger picture. Operating in this limited space and not so much worrying about the societal context or implications is deemed to be “professional”, doing the job. There is danger in this attitude as it bears the potential of being exploited for some questionable ends (I am only doing my job). On the other side, most people may not even have the privilege considering the context in the choice of their work; they do their best to add value to society and earn their living, it might be a matter of simply bringing food on the table.
- Quite often the message in commercials on TV and cinema these days seems to be: save the planet by buying this enormous battery powered electric SUV showing people and their smug faces. This message is akin to “consume your way out of overconsumption”, “snack your way out of obesity”, “drink your way out of alcohol addiction”, or quoting George Carlin: “Fighting for peace is like screwing for virginity.”
- Cars are even getting bigger (autobesity — hypertrophia automobilis), they are becoming living rooms with armchairs on wheels (SUVs — suvitis vanitatis) with higher resource demand when natural resources are diminishing, a mind-boggling aberration of society; providing more comfort and luxury is all about convenience (see The Cost of Comfort, Convenience and Complaisant Consumerism).
- There is a balance to be had between what customers want (and are prepared to pay for) and what they need. Businesses will satisfy wants by selling goods and services in order to make profit; often, they care less about long term consequences on both customers and the environment; this is where internalising externalities comes in as a feedback function to be performed by society as markets would fail.
- Urban planning includes urbanisation, i.e. the elimination of urban sprawl and concentration of population at higher, more efficient densities. This does not mean tower blocks, but thriving communities with integrated amenities.
- Spending effort on automated driving at the current state of affairs in the world is like worrying about the dinner menu on the Titanic: a nice problem to have and to solve if there is no immanent danger.
