Autonomous Cars: A Developing Challenge

15 minute read ★

“There are always those who argue that government should stay out of free enterprise entirely, but I think most Americans would agree we still need rules to keep our air and water clean, and our food and medicine safe. That’s the general principle here. What’s more, the quickest way to slam the brakes on innovation is for the public to lose confidence in the safety of new technologies.”

Barack Obama
44th President of The United States of America

The training and testing of this car is a common site on campus.

Obama’s address while introducing regulations for self-driving cars in American roads covers a fundamental issue often overlooked in the hype and noise of self-driving cars – the legislative, sociological, economic and political outfall inevitable with such advances.

The fully autonomous vehicle has been a partially conceptualised pipe dream living in the realms of imagination and science fiction for the past century. From James Bond’s Aston Martin to Batman’s Batmobile, if a car isn’t autonomous, it is not futuristic enough. Now fast forward to the 21st century where we have ABS, cruise control, auto-transmission and driver assist technologies present in almost 80% of the cars that roll out of production lines all over the world. And these technologies are such an integral part of transportation today that we do not consider them to be consequential features to be marvelled at anymore. The aforementioned technologies comprise what is more popularly known as the Level 1 of autonomy 1 in the field of autonomous technologies.

And we now exist in a time where the possibility of road legal autonomous vehicles within the next 20 years is a legitimate possibility; with some pundits claiming a figure of 10, or even 5 years from today. But the road to this dream is riddled with potholes: both the practical hindrances of actually getting such cars on the road, and the associated (presumably) wide-ranging socioeconomic changes associated with the possibility of a society that considers vehicular autonomy to be the norm.

The Indian Scenario

“When driving, you should assume at all times that everyone around you is suffering from a severe mental disease that makes them either suicidal or homicidal.”

Neil Miller
Web Journalist

Getting cars to drive themselves in the real world is considered to be a generally hard task – the machines have to account for the inconsistency inherently associated with an imperfectly modeled world. An autonomous car has to take decisions continuously and anticipate the decisions of other agents on the road – artificial or human. This a rather tall task for systems which currently require tons of data for even basic pattern matching. The issue is amplified the closer one gets to India, because of the piss poor condition of Indian roads. Patterns of any form or shape are outliers rather than the norm – be it the potholes, the lack of proper marking of any form, the varied traffic, or the lack of roads altogether. On top of that, cows, camels, and squealing kids aren’t known to have a working knowledge of road regulations. Meanwhile, the mass of autos, buses, rickshaws, cycles, trucks, and the like aren’t particularly concerned about the existence of any such rule(s) either, consequently making the decision making problem harder.

And the issue is not just technical. India has economic issues. It has political underplay – a lot of it. There are social factors, biases, bases, and trends factoring heavily against the introduction of such vehicles on Indian roads. Situations to consider if such should eventually pass, including the effect on the common man – often in ways not immediately obvious.

An Issue of Implementation:

“India will be the last one to get autonomous cars! Have you see the way people drive here?”

Travis Kalanick
Former Uber Chief

There are some obvious problems with introducing autonomous vehicles in India – the lack of infrastructure, the economic situation, possible social backlash. There’s also the legal component – autonomous vehicles, as of the 25th of July 2017, have been outlawed by the Union Transport Minister. However, even excluding these issues which are better classifiable as non-technical and logistical , there exist plenty other implementation details that are hard to get right across the world, and (very obviously) especially in India.

First up, consider the effect of and on telecommunications. The sector is integral to the growth of the autonomous vehicle industry. There is bound to be considerable growth in data consumption as autonomous vehicles require a lot of metadata that can be easily fetched from the internet – road traffic data, accident information, location data, local mapping, and so on.

In India, this turns out to be a very major issue. Large parts of the country lack proper network coverage, and among those that do, very few places are equipped with the high speed bandwidth requirement for such vehicles. Even metropolitan cities are rather poorly serviced when compared against the global standard. Just compare the definition for broadband – in the USA, the minimum download speed required for a network connection to be classified as broadband is 25 Mbps. That same requirement in India is 2Mbps.

Moreover, there is bound to be considerable growth in data consumption for autonomous vehicles that will communicate over mobile networks (utilising decision swarms, for example), and it is necessary to ensure constant access to telecommunication networks for the functioning of these cars. An electrical grid blackout or weather disruption at a busy intersection may have fatal consequences for said vehicles. On a smaller scale, any weakening of the network across a grid at dead zones may cause smaller, but equally fatal accidents.

A lot of foundational work has to be done before we can begin to look at self driving vehicles as a reasonable possibility on our roads – and the sheer quantity that remains to be done casts the oft projected figure of 20 years for the future to arrive in a shadow of doubt.

“At some intersections, there will be a traffic signal. At times, there will also be electricity for the traffic signal.”

Neil Miller strikes again, the typical DASA way

The other issue is with Computer Vision, or how the cars see. Perception is a complex problem, inferring and understanding the 3D world the car resides in from a set of cameras is not as trivial as it may seem. This subsystem of an autonomous car is currently powered by long and complex pipelines employing complicated neural networks on data taken from multiple modalities. Real world aids in this regard for the algorithms are road markings and signage, along with a consistent set of rules that are largely followed.

Seems simple enough, except when these start being major factors even in developed First World countries, you know it’ll be worse in India. Several developers of autonomous vehicles have raised issues about the US’s lack of consistent signage across wide stretches of its roads. Lack of such forces automakers to rely more heavily on (possibly old, inaccurate) maps and invest into more expensive, sophisticated sensors and technology. Errors here are crippling and can be fatal: as in the case of the first known death caused by a self driving car – when the car sensors of a Tesla Model S failed to distinguish a large white 18 wheel truck against a bright spring sky.

“People in Faridabad don’t drive like the people in Frankfurt! Same goes for Calcutta and California or Mumbai and Miami”

Anurag Ghosh
We couldn’t find anyone else

And then we get to India. We’ve waxed enough about the situation of Indian roads – rather, the lack thereof – and the problem is trivially obvious. Current research datasets like KITTI are inadequate for training autonomous vehicles for India for precisely one reason – the road conditions are so different from US and European conditions. And this is only considering the perception part of the pipeline, the behavior of people who drive cars in India is unlike say, the French. The Indian fixation of “jugaad” extends itself to driving on roads – laws are lax and considered as suggestions; people cut corners, drive on pavements, skip signals if the intersection is not crowded, unlike drivers in first world countries.

Another thing not to be left out of consideration is the matter of replacing our current fleet of vehicles entirely with autonomous cars. It is highly unlikely that all of the world’s, or even a locality’s vehicles will be replaced at once – not even if the governing authority demands so. It’s simply not a practical possibility, especially considering the economics of demand and supply and the current manufacturing throughput of conventional vehicles. What’s more likely is a progressive change. This implies years of a dichotomy wherein both self driving and human driven vehicles would take to the road.

The situation gets rather socially complex: assuming they stand the test of time and abuse, self driving cars shall go from a minority on the road to the overwhelming majority; from the hated section of the road to the one that is unequivocally “right” by popular consensus. It’s an interesting outcome to consider, if nothing else.

The Outfall – Economic:

“Today, you see millions of jobs are being created in the transport market by truckers and taxi aggregators. Such technology will make millions jobless. Maybe some years down the line we won’t be able to ignore it but, as of now… we shouldn’t allow it. The government platform will help get more people employment opportunities. The idea is in the primary stage but we’re working on it seriously,”

Nitin Gadkari,
Transport Minister of India
On why Autonomous Vehicles wouldn’t be allowed to ply on Indian Roads

Introducing driverless cars to an economy will naturally bring along with it a range of effects. A large number of them not that immediately obvious – for instance, it is reasonable to assume that the cost of the individual car would go up, but in a scenario with Level 5 autonomy 2, would owning a personal car make sense? In fact, when Uber was founded, co-founder Travis Kalanick’s goal was not just to create a transportation app but to make car ownership obsolete. Achieving that would entail driving down the cost of Uber rides so that they are eventually cheaper than personal car ownership. Currently the cost of ride hailing services such as Uber and Lyft are twice that of using one’s personal car. However, these economics are slated to dramatically reverse once driverless technologies become commonplace. It is entirely possible that owning a car may turn out not to be the economically sound thing to do in the face of hiring one from a service provider for periods of time; the economic shift may affect our ideas of personal ownership. If the situation seems to be a far cry, simply consider the Music, Television, and Movie industry: you may personally choose to buy all of it as it was conventionally sold, but does that really make sense to most in a world of Spotify, Netflix, and Amazon Prime subscriptions?

“The Indian government isn’t looking at the big picture when it puts the preservation of certain jobs ahead of the large-scale benefits of autonomous vehicles.”

Vivek Wadhwa
Distinguished Fellow and Researcher at Carnegie Mellon University

A potential fallout of autonomous vehicles often brought up is the matter of unemployment of labour as can be seen when considering those who drive for trucks and taxis. Among the first vehicles to be widely automated are these, in various countries around the world – as in Germany. Take away these professions, and you take away for many their only source of income, even more so because that particular skill set (driving) is overpopulated with a shrinking demand, as more and more people drive themselves instead of hiring drivers, and even this would be replaced by the cars driving themselves in this possible future being discussed. However, this is not that a trend that holds worldwide – in places like the U.S. for example, the trucking community is faced with a shortage of divers with the average age of drivers being around fifty. Introduction of autonomous vehicles would actually help the transportation industry there. This is however not the case in India where people are dying for any job which will fetch them two square meals a day, despite long hours and job hazards.

A common response to the job wipe-out claims is that job loss is standard with every introduction of ground-breaking technology. One could consider the example of the Luddites of Britain in the early 19th century. As MIT Economist Erik Brynjolfsson said, “A lot of skilled artisans did lose their jobs,” but several decades later demand for labor rose as new job categories emerged, like office work. “Average wages have been increasing for the past 200 years,” he notes. “The machines were creating wealth!” Essentially, our view is rather limited when it comes to employment in case of technological revolution, and as with that situation, here too it is not possible to predict the type and number of job opportunities the AI revolution would bring along with it. It’s entirely possible that the number of jobs scoped out by the technology more than accounts for the unemployment outfall. However, if that doesn’t turn out to be true, governments would need to turn to alternate options like Universal Basic Income (mentioned in a previous Eye to the Future article; Affording Unemployment for All, Ping Nov ‘16).

All said, such an argument does not account for a rather important factor: job skill mismatch. Consider Excavator operators: it’s a skilled job requiring specialised training that will be of no use anywhere else if machines take over the job; and machines are taking over – they’re just better. The market has jobs for those who lose their employment, but does it have a job that fits? The new employment opportunities opened by autonomous vehicles may require either a higher, lower, or entirely different skill set from that possessed by those humans displaced by these technologies and reskilling may itself be a hard problem to scale for such a large population.

The Outfall – Legal:

Bringing a new class of vehicles onto the road comes with a baggage of new legislation. Legislation that surrounds the matters of personal property, liability, insurance and safety measures. Different places have different laws regarding how vehicles should be driven – their speed, pollution level limit, lane usage, overtake laws, and the like. In the current situation, most of these aren’t particularly followed, but an autonomous vehicle would be expected to do so to the dot as it can be “programmed”. But given a situation where a vehicle does follow all of these rules and still ends up in an accident or any other such predicament, the legal system would have to come up with methods to deal with it.

Consider matters of liability. When autonomous vehicles get involved in accidents, the issue of liability may get complicated as resolving the question of fault will require consideration of novel and challenging questions, considering that the “driver” of the car is no longer human, but rather software programmed by a lot of people working in tandem with other people writing more software or designing hardware for dedicated tasks. If one cog in the machine goes wrong, the natural instinct is to blame that individual responsible for the cog, but such judgements get complicated when it’s a team working on something, possibly relying on work other people did.

A direct example of this would be seen in the effect on Insurance premiums. The extant test is to determine the driver’s liability and adjudicate accordingly, but in cases where there is no driver and the car runs entirely with software assist, there need to be different parameters of consideration. Traditionally, the underwriting criteria for insurance companies depended on the number and kind of accidents an applicant has had, the miles they expected to drive, and the place where the car is to be garaged. While some of these criteria would continue to apply, the make, model, style and the manufacturer of the car may assume greater importance. In the driverless future, actuaries may have to replace calculations about individuals with issues such as hacking of cars, analyzing which parts of the country have better satellite imagery etc. They will also have to identify the difference in quality of the safety features across driverless cars.

In fact, given that autonomous vehicles promise to improve safety standards manifold, one may have to pay a much higher premium if one chooses to drive cars on their own rather than let the computers take over. Another question is how much should be left to the commuter’s will? Manual drivers may genuinely be unable to conform to required standards temporarily, but as long as they’re on the road in such a state of nonconformity, they would be considered a threat to everyone else.

“We think this is a really critical step towards imposing and working towards the level of safety needed in this arena”

Anthony Foxx
U.S. Secretary of Transportation
After issuing guidelines for self-driving cars

In order to reduce legislative complexity and to ensure commuter safety, it is expected that there would also be a base standard and standardisation body/process for systems performance and care that all vehicles should conform to. It would presumably be managed by a government regulatory body, although it is possible that self regulating bodies may arise as in the case of the Central Board of Film Certification in India.

Sadly, such a discussion is more likely than not futile in India, where politics controls all. The total number of bills pending in the parliament is around 28 and 354 have lapsed because they hadn’t been passed by the previous governments. Some of the Bills include the ones like The National Road Safety and Traffic Management Board Bill, 2010 and The Consumer Protection Bill, 2018. So if such important bills like the one mentioned above have lapsed then it is very possible that India has a long way to go before a legislation is even voted on in the Parliament regarding the deployment of autonomous vehicles on Indian roads. The situation in Courts of Law isn’t much better. The top five tribunals in the nation have a combined backlog of over 3.5 lakh cases.

The Outfall – Social:

A major hindrance to introducing autonomous vehicles anywhere at all is social response. The introduction of such autonomous might not be well accepted by the multitude of the Indian population, for instance. Firstly we have commercial vehicle drivers and owners losing their jobs. And here we are talking about numbers not in hundreds or thousands but millions which is about the population of a small European nation. Disgruntled masses may take to the streets and the legislation is bound to be powerless against such a humongous force. Even if these people are provided with alternate jobs, there will be the problem of ‘job skill mismatch’ which we have already discussed above. Will these people be happy with their new jobs? For a country whose unofficial motto has been, ‘chalta hai, chalne do’ (if it works, let it work), it may be hard to adapt to such a tremendous change. The natural disbelief in new and forward technology is bound to make a lot of people uncomfortable with the idea of subscription.

Leave alone subscriptions, people simply do not trust machines as much as they do humans, even if the machines are known and shown to be the safer option, and this is not just restricted to India. This effect is particularly prevalent at the onset of any new technological breakthrough that requires extensive testing to better itself, which is the stage autonomous cars are at. The issue faced by these in particular is that any mistake could be fatal, which strengthens people’s mistrust of self driving cars – thus reducing the scope for free testing of the cars, hence limiting the path to improvement. It’s a vicious cycle that either requires the cars to be near-perfect from the get-go, or for people to abandon their apprehensions. This behaviour to autonomous vehicles isn’t just a response to their performance, but also in the decision making process involved. Given a societal problem with a widely accepted solution, society has greater tolerance for human failures. Wade into genuine ethical problems, and the water around computerized decision processes get murkier.

The Trolley Problem (and its variants) is a common ethical dilemma often brought up in such discussions. For those unfamiliar, it (the specific problem here) goes such: an autonomous vehicle is headed for an unavoidable collision, and there are two ways to deal with it – either kill the other party involved in the collision, or kill the car occupants. It must choose one of the two, there is no way out.

We leave the actual ethics of either action up to the reader (there are many considerations one can make, and a Facebook meme page dedicated to this), and ask them to consider the possible consequence in implementation. The government may incline towards regulations in favour of making the decision that would “cost the least lives”. However, they may choose to leave it up to the manufacturer whether or not a vehicle should try to protect its own occupants or minimize fatalities (should the two goals be at conflict), with incentives to those corporations that prioritize minimizing the damage. Which means manufacturers could sell specific vehicles, or dedicate their entire fleet to vehicles that prioritize their occupants’ lives over the “greater (heuristic) good” – for a premium paid by the consumer to account for the excess charged by the government. Consider the effect this has on society – one would essentially be able to explicitly buy themselves their lives at the potential cost of others’.

There are other social and cultural issues if we look beyond the ethics of automation. If autonomous vehicles become the norm, leading to the decline of manually driven cars, it would also affect our psyche and culture. For example, how can we compensate for the denial of joy of driving to a life-long car enthusiast? So bid your goodbyes to the American quarter mile drag or the Tokyo drift because now your car is built neither to let you drive nor to drift, unless you are at a commercial “car stable”, driving cars like they are horses.

Given the direction the vehicular industry seems to be heading regarding autonomous vehicles and personal vehicular ownership, it’s possible that auto manufacturers and other companies like Uber would follow a subscription based model much akin to Netflix or Spotify today, where the vehicles would be rented. From a legislative point of view, this simplifies matters by a lot – since the brunt of vehicular upkeep is on the managing company, all but a few cases could be attributed to them. Despite the simplification of legal liabilities because of decline in individual ownership, it would come however, at a cost of individual identity and liberty – a topic that might be a touchy subject to some, for reasons ideological and otherwise. For instance, Vintage car communities might flounder, as they wouldn’t be able to buy and drive their vehicles on roads and convene for events, which many would consider a huge cultural loss.

The Road Ahead

Given the current status on autonomous vehicles with respect to their technological development, it’s highly unlikely that they shall be street legal within the next five, or even ten years anywhere in the world – and when considering India, twenty seems to be a safer bet. But while the technology may or may not progress, the question remains as to whether the world will be able to adapt quickly enough to accommodate for the inevitable changes.

Levels of Automation:

There are in fact six levels (scale of 0 to 5) of car autonomy from “no automation” to “full automation” as defined by the SAE International (Society of Automotive Engineers). The ones mentioned in this article are:

Level 1 (Driver Assistance): Vehicles with this level of autonomy, in some driving modes, can handle steering or throttle and braking – but never both. However, the driver must be ready to take over those functions if called upon by the vehicle.

Level 1 cars also have some systems that use information about the driving environment, but the human driver monitors the driving environment. Level 1 autonomous systems have been available on production cars for several years, and features such as self parking and lane assistance fall into this bracket.

Level 5 (Fully Autonomous): At this level, the vehicle needs no human control at all. It doesn’t need to have pedals, or a steering wheel, or even a human onboard.

The car is fully automated and can do all driving tasks on any road, under any conditions, whether there’s a human on board or not.

This article is part of the column ‘Eye To The Future’ that is dedicated to long form articles based on contemporary and future impact of technology and scientific impact on society.

Call for pitches: Interested in contributing to this column? Shoot an email to ping@students.iiit.ac.in with your pitch for the column.