What is the final destination of autonomous vehicles?

Are we ready for it, as Taylor Swift might ask...

At the end of last month, the British Computer Society held a talk on whether the big “we” (humans) are ready for autonomous vehicles. Given that autonomous vehicles have already generated some pretty meaty IP disputes (trade secrets misuse, patent infringement, from Silicon Valley to China), the AmeriKat asked Luke Maunder (Bristows)  to report on what the technologists had to say.  Over to Luke: 

"Some, including this guest Kat, were attending having only recently watched the new X-Files episode in which artificial intelligence, including a self-driving car, had learnt its behaviour from considering how humanity behaved. It wasn’t pretty. Much more seriously, however, all were attending under the cloud of the recent tragic death in Arizona, when a woman was hit by a self-driving car and lost her life. This was clearly at the forefront of the attendees’ minds during the talks, especially when one was looking at the intertwined issues of safety and consumer trust. 

A great balance was struck by the speakers between the optimism for a better tomorrow, the reality check of the practicalities and the cynicism of whether there is actually a demand for autonomous vehicles. Ajit Jaokar, the course leader on Data Science for the Internet of Things at Oxford University, acted as a moderator. His view was that either we pursue this in the UK, or let other nations get ahead and set the terms. I can’t say I disagree, and at the forefront of all our minds must be the inevitable trove of IPR generated off the back of any autonomous vehicle research, which will inevitably have relevance to other machine learning and robotics applications. 

The story to date 

Claudio Mutiis, a data scientist and AI researcher, began with an explanation of the development of autonomous vehicles, starting in the 1980s with Dickmann’s vision-guided Mercedes van running up and down the autobahn, through the DARPA Grand Challenges and then onto more modern developments. 

Views from the BCS event

Everyone familiar with this area will know the five levels of automation. Whilst it seems like progress has been huge, in reality we are presently transitioning from level 1 to level 2, with key automated capabilities being deployed (for example, my car automatically braking to prevent a “fender-bender” when the car in front slammed on the brakes unexpectedly), but where humans are very much still in the driving seat. Tesla and Audi are pushing the boundaries of level 3 with their latest vehicles, but whilst levels 1 and 2 have been what this site would call kitten steps, level 3 might need something bigger.

Of course, in line with recent events, one must also consider the ethical issues and how the industry can be regulated from that perspective. This is such a complex question it cannot be done any justice in this article, so I will not try. I will merely content myself with encouraging readers to visit MIT’s Moral Machine, which asks you to solve the very difficult questions that machine intelligence might need to consider. Once you have provided 13 answers it gives you a report on where your priorities lie—self-illuminating, if nothing else. 

What was clear from Claudio’s talk is that if one removes the economics question, the challenges in self-driving cars are very much software driven. In his view, the hardware is broadly there and, but for the inevitable need for faster processing and data communications, software is where the time is being spent. As progress marches onwards (as it inevitably does), the importance of patenting software driven inventions and copyright protection for code only become more apparent. 

Are we sure about the hardware?

Maximillian Odendahl, the Co-Founder and CEO at Silexica, then set about dispelling any rumour that the hardware is broadly there. Clearly a vast amount of processing power is required to manage the data being generated from multiple instrument clusters in real time. It is not just about processing that data, but making sense of the combinations, taking the best of each instrument while side-lining its particular weaknesses.

With multi-core chipsets reaching a plateau the answer lies instead in refinement and optimisation. In particular, using specific processors (CPUs, GPUs, etc) to process specific data types or algorithms that they have been customised to process efficiently, and breaking down software execution to run in the most efficient manner possible across multi-type multi-core chips. 

As chipset companies work on these solutions, the question arises of how to protect such developments. Patents for the hardware or technical solutions found in software, as well as copyright in the code, have already been touched on, but it is also worth mentioning trade secrets. Particularly when it comes to optimisation routines and algorithms, keeping them as trade secrets might allow an edge to be preserved over the competition that would otherwise be lost. As the world becomes more inter-connected and there is the need to regulate and involve Government in preparing the infrastructure, one can hear the siren call of standard setting and FRAND licensing on the horizon (indeed some patent pools already have licensing programs). Any trade secrets held back can be the secret sauce by which one stays ahead of the competition, as long as they remain secret. 

What is the reality now?

Next stepped up Grahame Bennett, head of one of the largest vehicle fleets in the UK, belonging to the Royal Mail. He gave the perspective from industry, concentrating on the Royal Mail’s own R&D on the topic, and his own experiences. This was the first time that the question of how autonomous vehicles might be used day-to-day really came up, with a distinct line being drawn between, for example, vehicles for performing repeat routine functions in a busy geo-fenced yard, lorry chains on motorways and all of us enjoying being driven by the car to our local for Sunday lunch. An interesting perspective from one of the people in the UK you would be trying to sell your technology to; needless to say, the use cases needed to be justified.

Now what?

The five levels...

Charlie Henderson, a partner at PA Consulting Group, gave the final talk on how, with the technology being here and the regulations being in place, it might not be “if” but “when”. However, out came the figurative black cap as we were challenged by Charlie on “why”. The focus here was on consumer demand and on the realities of generating interest and investment in the technology necessary to reach level 5.

The ethical issues were addressed briefly, but were unlikely to be the key question. As an exercise in ethical philosophy they were interesting and crucial, but when the time came to make a call, the answer was likely to be what made the occupants of the autonomous car feel safest; otherwise, who would get in them.

What was key in practical terms is how reality interfaces with the dream. He provided a number of examples, but the starkest one for me was how city centres might fair. Take London, for example, where “jaywalking” is not an offence; indeed, it is a fact of life. Now imagine how the knowledge that every car will, in all but the most extreme of circumstances, stop the moment it detects a pedestrian or a cyclist in front of it. Welcome to the newest car park, as the vehicles certainly aren’t going to be moving around. This is simply another problem to be solved, but where an obvious solution is regulation, rather than technical, for example to introduce jaywalking as an offence to the UK, public perception follows.

Separately, there is the basic fact that people often enjoy driving. I still miss the MX-5 I gave up when my family came along (granted it was an easy choice, but one can have fond memories as opposed to regrets). On Twitter after the event, Dr Brigit Clark taught me a new word: Fahrvergnügen, or the pleasure and joy you get out of driving a car. The point is trite, but what is the future of any fun or enjoyable car in a totally self-driven future?

Charlie Henderson concluded with a simple thought, to paraphrase: autonomous vehicles aren’t coming, they are here, but while that ship has sailed and they will now be staying, it is questionable whether they will ever be in every garage."