Automated driving: The search for perfection

Part one: Prof. Nick Reed, Reed Mobility

It’s a debate that has been raging for the best part of two decades – when will autonomous vehicles be a commonplace sight on our roads? In the 1990s it was thought that by 2010 automated driving would be de rigueur as auto giants such as DaimlerChrysler and Ford were investing vast sums of money into their driverless vehicle programs.

And then 2010 came and went. Despite some extraordinary advances in technology that had seen the AV sector make huge strides there was still a lot of speculation that perhaps those predictions from the end of the previous century might have been a little bit too ambitious after all. And then Google and subsequently Uber entered the fray and the landscape shifted again, this time quite seismically.

Skipping over the intervening years, with thanks to the notion of artistic licence, we’re now in the third decade of the 21st century and hurtling towards the “new” date of 2030 that experts the world over started to focus on as a realistic target for the widespread implementation of autonomous vehicles at the beginning of the last decade.

Challenges for automated driving

So what’s the hold up? Read Intertraffic’s interview with AV guru Richard Bishop, published in December 2020 and it’s evident that the trucking sector is edging ever closer to being the first automotive sector to make a game-changing breakthrough but how much further ahead is it? What is stopping one of the car behemoths from making the kind of announcement that we’ve been expecting for 25 years?

Or is that oversimplifying a very complex issue? UK-based autonomous vehicle expert Professor Nick Reed, formerly of TRL but now the CEO of Reed Mobility, believes that it’s a combination of factors. “I usually think of there being three barriers. One is legal, so making sure we have the kind of regulatory framework in place to accommodate automated driving,” he says. “And I think one of the critical elements in achieving that is being able to record the right data and access the right data to be able to say how the vehicle was behaving, was that acceptable and who should be considered responsible for the actions of the vehicle.

“The second barrier is technical, It’s still not perfect but we know exactly what the right combination of sensors is. Are they reliable in all situations where we expect them to perform in all environmental conditions, road conditions, and so on?” Reed, one of the mainstays of the UK’s Meridian AV project that was launched in the London Borough of Greenwich in 2015, concludes his triptych of potential barriers. “The third one is societal. Do we know how we want automated vehicles to integrate into our lives and can that happen in a way that is commercially viable?

So I think across those three, legal, technical and societal, there can be a tendency for one to blame the other two for not making the progress that that was expected. But I think there are challenges across each of those domains. There's been a lot of progress in the movement of goods, I think we're seeing some commercial deployments of automated vehicles performing that task. But certainly on the on the passenger side, it could be said that progress hasn't been quite what was expected 10 years ago.”

AVs: Being for the benefit of everyone

Reed’s mention of society brought about a memory of a brief conversation that this author had with an acquaintance from an entirely different sector only a few weeks ago. “What is the main societal benefit of taking control of the task of driving away from humans and giving it to essentially, algorithms?” In other words: what is the main societal benefit of automating driving?

“What got me into the field was safety,” he replies instantly. “I've done years of work at TRL on driver behavior, the effects of our phones, the effects of drugs, alcohol, fatigue, and I always felt I was the one saying you don't want to do that you want to do this. And when I thought about automated driving it felt a more positive way to tackle the issue. This would enable safer road use and free up time for other productive or social uses.

So that was why I started working on automated driving, it's the sense that it would tackle some of those contributors to road crashes that humans are sadly susceptible to.” Reed continues: “Another societal benefit that is overlooked or underestimated is the extent to which automated driving might give us better management of the use of the road. We can get increased capacity, and we can manage capacity more effectively when we know who's traveling where and when and have control over how that all proceeds. I think that across safety and management got the inclusion factor as well so that it would enable independent mobility for elderly or disabled travelers for a greater period of their lives. I think that's an important aspect as well.”

In the balance: the AV/human-driven mix

One area of concern that has been a constant in the realms of AV discussions over the years has been “the mix”, or rather, when will the amount of autonomous vehicles on our streets start to cause an issue? It’s a delicate balance alright but when does it become… dangerous? The amount of vehicles being driven imperfectly by human beings mixing with vehicles impeccably driving themselves, observing rules, has legal experts running for the hills.

“That’s really good question,” nods Reed. “Who gets to define what perfect is? There have been studies that have shown you that AVs are often responsible for causing collisions by stopping in places where drivers weren't expecting them to. I've been working with the British Standards Institute (BSI) on a digital commentary driving concept whereby we collect data from vehicles that tell you what they were seeing, what they had perceived, how they were predicting those objects would move and how were they planning their trajectory in response to that. We are also looking at if that should be accessible and shared – albeit not publicly, but at least to a regulator.

“It would be for every AV developer to produce this data, should it be required. I think once we can get to that we can start to dig into how AVs are behaving… and is that how we want them to behave? Is that how they should behave?”

Levels of autonomy

It’s been several years since the Society of Automotive Engineers’ (SAE’s), six levels of automation were drawn up (level 0 = no autonomy, level 5 = full autonomy. A lot has changed in terms of the development of autonomous vehicles – a lot of autonomous water has passed under the bridge, so what does Reed think about the validity of those driving levels, especially with the recent advances in truck autonomy. Are those five levels of autonomy still relevant and fit for purpose? Do they need to be re-thought?

“From an engineering perspective, where they can be challenged, is how they are then used when you get to commercial products, which might have different levels of automation at different points within a journey. How is the vehicle then described to the consumer? And do they understand that description? And do they then use the automation in the way that it's intended? So I have no issue with the SAE levels, per se.”

“There are very smart people that are thinking about what they entail and what they represent. So I think they're as good a description as any but when you think about level four, which could be what we think of as a normal car that does highway driving versus the Waymo vehicle that does all the driving all the time, but only in a very restricted area. They're both level four, and maybe there should be some distinction between those in terms of how they're understood by a wider audience.”

All about marketing

The wider audience – a key target market for anyone launching pretty much anything, be it a soft drink, a shoe or the concept of the driverless car. Public perception is paramount but it all hangs on how the message regarding its usefulness is conveyed. ‘I must have one of those’ is the end goal. Do autonomous vehicles need the same kind of marketing push? Where is the “you must have one” campaign going to spring from and how, as an industry, can the notion that driverless cars are not safe be put to bed once and for all?

“People will need to be reassured that they're safe, especially when you are effectively taking control away from the human operator,” agrees Reed.“I saw some statistics where something like eight out of 10 people think they're above average drivers. But my experience of trialing these vehicles is that people become comfortable with it very quickly. If it means that they can get on and do other things and watch Netflix or chat to their friends, or have their dinner or do their makeup or whatever then I think people will become comfortable with it. And we already see it today with the misuse of semi- or partial automation systems where people are only too keen to claim back that time and use the system in ways that they shouldn't be. That's the risk,” Reed elucidates. “When we get to a situation where there are cars capable of doing that, do those behaviors then get carried across into cars where that shouldn't be done.”

The mention of safety leads Reed to refer to the death of Elaine Hertzberg, the first pedestrian to be killed by an autonomous vehicle, in Phoenix, Arizona, in March 2018. The whole AV sector could very well have been hugely negatively affected by the accident, coming at a time when AV safety was the subject of much discussion.

“The car had detected the presence of a person, but it didn't know how to respond. The safety driver was there to help manage that situation. But they were being distracted by their smartphone. Even in that situation where you had a professional safety driver involved there’s a misunderstanding of the capability of the vehicle because it had done the right thing so many times before, they weren't expecting to see a pedestrian at that point. And suddenly there is one right upon you when you should have been paying attention…and very sadly the worst happens.”

Autonomous trucks: a long weigh to go

As previously highlighted, the AV world’s current flag bearer is the trucking sector. Innovators such as TuSimple and Embark are among those leading the race to deploy driverless freight solutions in the coming years, with the obvious markets of trans-Continental goods deliveries and mining just at the forefront of their brilliant thinking. As far back as December 2019 an autonomous semi-truck drove itself across the US to deliver butter in near-perfect condition and things have moved on at a rapid pace since (in spite of the pandemic, and some would argue because of).

“The butter run is a great example of hub-to-hub automation,” concurs Reed. “It’s a simplified driving task, it's highway driving where all the traffic is traveling in the same direction, you've not got junctions you've not got, or shouldn't have, pedestrians, there’s no cyclists, so the automation task is simplified, and therefore you've got a strong commercial driver for automation. I've done some work on rail automation as well with the automated mining trains in Australia. You have these 44,000-ton trains, taking your mined materials from the mines to the docks. One of the main drivers for implementing automation was with a human-driven train they needed to stop to change driver halfway along the journey. This means that you've got to slow down a 44,000-ton train and then accelerate it again and that's hugely expensive and time consuming.”

However, as Reed explains, it’s not just a safety issue in play. “It isn’t solely being able to remove the driver from the situation, it's also the operating fees that you save through automated driving so that you don't have to stop every four hours. You can operate the train, or in our example the truck, 24 hours a day. Trucking is a great application for automated driving and I must say that Richard Bishop is a great person to be pushing that forward.”

Where we are versus where we could be

“To be honest I thought there would be commercial deployments of AV's in cities by now. I thought that was realistic and I still think that it was. There was a sense at the time, 2015-2016, that we’d completed 95% of the task and that there was just this little bit to go and we'd be there - without realizing that that last 5% was as hard as the previous 95. In real terms it was 50% of the way there,” he admits.

“I think this is why it's taken as long as it has to get to where we are. The technical challenges have been harder - the Law Commission have been doing fantastic work on setting out the regulatory framework so that's coming together now. The DfT [UK Department for Transport] is also doing good work on how it will regulate and approve automated vehicles. I think the technology for automated highway driving is basically there. It's the level three systems that are coming to the market now. Then there are automated shuttles, low speed stuff for passenger and goods deliveries, and they are getting there as well. We are starting to see automated vehicles in specific use cases and over time and those will grow and become more sophisticated and be able to cover more of the driving tasks that humans have to complete today.” Automated driving has meant that experts such as Prof Nick Reed have to question virtually every aspect of the term “driving.”

“Automation has caused us to ask ourselves what is a road? What is a driver? What is a vehicle? All these things are basic fundamentals that feel very concrete to us in the old system. But now our thinking is different. Do we count a computer system as a driver? Or does it have to be called something different? Should that term that be reserved for a human operator of a vehicle? We’re now coming to a kind of universal agreement over what the terms should be and what they mean.”

Is it fair to say that automated driving is at something of a crossroads? If so, which way is it going to go? In part two of this feature Intertraffic talks autonomous vehicle roadmaps with ACEA’s Smart Mobility Director Joost Vantomme.