SoundCloud Widget Placeholderhttps://soundcloud.com/uclsound/episode-2-problems-and-potential-in-self...
Episode 2: Problems and Potential in Self-Driving Vehicles
Hello and welcome to IM@UCL: The Podcast, a podcast about the research at UCL that will revolutionise the future of driving. My name is Cassidy Martin, and I am your host on this journey of self-driving discovery.
As humans continue to test their limits of potential with technological advances, problems sometimes arise. Take social media for instance, it's a great way to stay connected with friends that are far away, you can use it to organise events, self-promote your business, and so many other things. But social media has its problems too. Users accounts get hacked all the time, sometimes scams are pulled off with fake accounts, cyberbullying is a huge issue for younger users, and users’ data is collected and sold off to companies. This is why IM@UCL is approaching advances in self-driving capabilities with extreme caution, taking every opportunity they can to safeguard against potential issues.
For this month's episode, I spoke with two researchers who will be contributing to IM@UCL’s product design safety. The first is a researcher who uses mathematical modelling to solve any systems issue, and another who specialises in how people's behaviour and performance are affected by lighting.
Let's get started.
My first guest is a very unique type of engineer.
I'm Francesca Boem. I am a Lecturer in control systems in the Electrical and Electronic Engineering Department at UCL. And my research activities are in the area of control engineering, and more specifically, I develop methods to detect anomalies and faults.
And this method of detecting anomalies and faults can not only be used for mechanical systems, but…
…industrial, plants, and things like that. But we can think also natural systems. So, we could describe as a system also, the behaviour of some ecological systems or parts of the body can be described and modelled by via dynamical system.
The steps taken for this method, regardless of the system are as follows:
When there is a new problem, so a new application, what I need to do is so first of all, to understand the problem. And there, you don't need technical, let's say, skills. Or not, the engineering skills. You really need to try and understand in detail or in depth, where the problems are, where are the main elements. And then I need to, to build a model. So, a mathematical model that can describe the reality of the physical problem or to translate things from one language to another. And when I find the mathematical model that that is well suited to explain what the problem is, then I work on the mathematical model to find solutions.
When it comes to IM@UCL’s self-driving vehicle designs, finding potential faults and anomalies, and solutions for them, is essential for ensuring designs are safe for use. And this is exactly how Francesca intends to contribute to the facility.
We want that interaction between the person and the car does not create problems for the safety of the person. And therefore, I could work mainly in understanding if there is something which is going not as we were predicting. So, to detect if there is any fault in the car in the autonomous car. But also, to make sure that this feedback that there is between the car and the driver, and actually also it's two ways feedback in that sense, we want this does not create problems for the functioning of the vehicle. Then my research works also in the interaction of the car with the rest of the world. And in this sense, the car can be connected to other vehicles or to, for example, some systems that give instructions or give guidance, for example, for traffic control and things like that. In this sense, the cyber physical security aspect is quite important because the vehicle is not on its own. But actually, it's interacting with the rest of the world. And we need to be sure that there are not problems in the interaction between the vehicles. So somehow, we need to optimise the behaviour of the different vehicles which are around the car. But also, that there are no problems coming from the external world. As I said, it could be cyber-attacks. And then finally, my research can actually be used also on the optimisation of the network, traffic network. So, if we don't consider the single vehicle anymore, but we are considering actually a network, which could represent an entire town or a part of it. And we want to optimise the situation in terms of traffic, we want to reduce pollution, we want to reduce congestion, and so on. In that case, the network itself could be seen as a system. And it's really interesting to use some of the control matters, to predict how the network will be in the future, based on what we are seeing, at a certain point and from the history that we have. But of course, any also any advice that we can give, for example, we could say we could advise the cars to follow certain paths, any advice will create a change in the network. And we need to take into account that this change, actually will, will of course affect the status of the network in the future.
Yeah, I think that's such a complicated thing. Because you want to be able to gather all this information from this network of cars so you'll be able to drive is safely and as environmentally friendly as possible, I guess. Making sure you’re creating as little footprint as possible, but then at the same time, like trying to make sure you're mitigating having security hacks, and all that kind of stuff as well.
Yeah, I mean, as I said, my research is really more of a support for the other people, to enable other technologies and to help the decision process of other disciplines. So, in this sense, it's quite a background discipline that can support different activities, and it can be useful for to solve different problems. Yeah, it might sound that like, I can do everything here. I mean, it's not like I have the solution to all the problems that we have discussed. And that's also why I really would like to find some specific problems where I can give a contribution and where there could be shorter term factors. So, it's, it's really about finding where the knowledge that I have, the tools that I know, where they can make the difference in the systems that also the other researchers are developing.
So, Francesca cannot solve all the problems at once, though, she can solve a number of specific problems. But there is a particular challenge when it comes to using mathematical modelling to solve problems in this particular context.
So, it's not only uncertainties related to the environment, the fact that you can have different things happening around the vehicle, you have different things happening on the traffic network, but also the fact that there is the human in the middle in this system. So, there is somehow some subjectivity, there are some psychological aspects, there is sometimes the unpredictability of the human person. So, it's not a system like others. Even if we are trying to make it autonomous and actually removing somehow the human action, but since the human is anyway involved, so it will be quite interesting and challenging to consider this element in my analysis, which usually it's not the case for most of the systems that I have considered in my past research.
Yeah, that's true because you are. You're having to think about not only the environment, but how a person is going to react, and people are unpredictable. So, you can only, there's only so much that you can, that you can do I felt like with that. Well, I'm not saying obviously there's a lot, but I mean like that, yeah, there's always going to be something that you can't account for I felt like.
Yes, sure. It's a bit, yeah, there is some form of unpredictability that you need to deal with somehow. And it might not be possible to have 100% guarantee that everything will go as you wanted.
Francesca wants to develop methods to guarantee safety and security of self-driving systems, even when things do not go as predicted. And although this may prove to be quite difficult, she is up for the challenge and looks forward to the opportunity to collaborate across departments.
I think that this project will give us the opportunity of some early career researchers to really meet and build this synergy between the different disciplines. I think something new can come up, both in terms of problems and potential. So, I'm really looking forward about this synergy. The fact that it's not only about applying my research to a problem but finding really where the different disciplines meet and how this can build something new.
Francesca's research may be more of a background discipline. But its impact on the self-driving field and IM@UCL in particular will be huge. My next guest is another team member of IM@UCL who will be looking at problems and potential in self-driving but in a different light.
My name is Jemima Unwin Teji. I am a lecturer and programme leader on the MSc Light and Lighting at UCL.
Jemima’s research includes a number of strands.
For example, how we judge a seen as effected by sound. So, we might think, ‘okay, in this cafe, we're happy to have it a bit darker and we have a certain type of auditory input’. Whereas we want it brighter if we have another type of sound. There's a lot of talk now about the effect of lighting on health and well-being and ultimately it comes down to daylight. And really the anywhere near to that is, you know, go outside. You’re never really going to get light inside that will be anywhere near to what it is out. You get outside in the morning, you've helped and trained your circadian clock. So that just means we have this natural bio rhythm, well lots of bio rhythms actually, and one of them is the sleep/wake cycle. So, if when we get up every morning and go outside in daylight, we reset our clock. So, I've done a bit of work looking at, you know, how that affects sleep patterns, although that has so far been inconclusive, mainly because there's so many variables. Then I've also done some research on pedestrian reassurance, so perceived safety after dark. So that involved looking at light patterns on streets. And generally, people didn't like dark patches. So looking at, you know, how dark you could get and how long those dark patches could get before people got really worried.
Now, Jemima will be applying her research to a new strand, self-driving vehicle safety. And she has a particular interest in looking at manual and self-driving in the dark.
The issue with autonomous vehicles is all the big tech players are investing massively in it at the moment is these intelligent systems. And well an autonomous system that's trained properly might not make the same mistakes humans do and they might not have accidents. And that's absolutely huge, because traffic accidents are quite rare in the West anyway. But when it does happen, it's absolutely catastrophic for those involved. So even if you only have a few a year, then you want to stop them because the impact massive for the family that that hits. So, I think then it's always the that 1% of instances where the machine can't predict it, then you need the human to step in. I think that's the challenge, isn't it? We know that car accidents, they do increase after dark. And that's when you look at the controls. So, if you're using the clock changes as the control period and you’re looking at the narrow windows before and after clock changes. So we have an interesting situation where you've got the same clock time, but different lighting conditions one week before one week after, before and after the clock change. So you can use that set of a very careful control where you look at are the accidents increasing in these narrow windows? And they are. And that could be because of lighting. Now, what will be really interesting will be to look at is that happening less in autonomous vehicles where the autonomous vehicle setting of actual autonomy is being used? Because then you could find out is it helping or not? Because if humans are making error in the dark and the cars aren't, then you could argue that it's safer, but we don't know that yet. So it might not be. There might even be more accidents after dark in an autonomous vehicle.
Finding out the answer to this can be complicated.
The problem is getting the data, isn't it? Because you got a car manufacturer that no one's going to want to hand over data, which says that their systems don't work quite well after dark. So, you know, you need to win these people over and work with them. And you know, they would put all sorts of caveats. So you know, they won't want you to say, ‘oh, actually, your car's not so safe after dark’. So getting data’s a challenge.
The good news is the advanced research facilities that were created at IM@UCL could really help make this type of research and other lighting research in semi-autonomous vehicles, possible.
But it would definitely be interesting to have a look to simulate a scene and have people drive in the simulator and see if, for example, if you also simulated somebody stepping out onto a road, could they see them more? Would they break sooner, you know, in certain lighting conditions than in others? What's important? Is it the background luminance? Is it what the person's wearing? Which would help differentiate them from the background? Is it their proximity? So, in certain lighting conditions, you can see them sooner? Because basically, that's what it's about, isn't it? You need to be able to see people before they step out on the road. Or is it that you know, most accidents are caused by people who are drunk and then lighting has, you know, very little impact then. So, I’m obviously not going to put drunk people in the simulator. Imagine getting the ethics approval for that! You're going to drink like, five, five pints. Not going to do that, are we?
That would be some fun research, I guess.
Yeah, yeah. But I think understanding how the way a scene is lit effects, you know, whether we can see people, or even obstacles on road, something's got on the road that could create an accident, tire or something. This could possibly all be simulated. Although, we may have the issue about dynamic range. So, we have very high dynamic range in a real situation. And we need to check that we can actually mimic that.
What does dynamic range mean?
So, it's like the range from very, very bright to very, very dark. So if you're looking at a screen, because it's a screen, its got a certain luminance, so it never gets really glaring, is it? Whereas in real life, when you're driving, you might sometimes think ‘I actually can't see’. That's called disability glare, or discomfort glare, where you can see but it's just not very comfortable. You know, like when you're driving in the rain and blobs of water around the windscreen. And then they kind of become pools of light and you’re like squinting. I don't want to run over a cat or a person, even worse. So that's the dynamic range. It’s basically that range that you get in reality, if that can be mimicked in the driving simulator, then it's potentially really, really valuable.
Yeah. And who would you say it's most valuable for?
Well, it's street lighting designers, engineers, everyone involved in design. And obviously if it's a potential, you know, if this research avoids an accident, then it's good for the general public as well. I think that's quite important. Obviously, researchers are interested in it. People involved in traffic safety, local authorities, TfL ¬– they might even be interested. Yeah, I would say local authorities, TfL, traffic safety, traffic management, people interested in those fields. Yeah, and architects, engineers, designers, you know, people involved in designing the built environment.
So you’d have a huge impact then. Like when you could potentially help a lot of different, a lot of different subjects and a lot of different people.
Yeah, but it's also hard to prove. So, you know, my worry with this is that you may get cracking with this. And then you end up with the kind of sad conclusion, which researchers are always guilty of which is, ‘oh, we need to do more research’. You know, ‘we've done all this research, and now we need to do a load more’. And then you know, because is that certainty, isn't it? You need to be certain, is the probability that what you're saying is not due to chance. And that's always quite challenging. You need to set up quite controls. But that's actually, that's another reason that driving simulators really useful is because it lets you control the variables which you can't do in real life. So, in real life, you need absolutely huge samples. Say you've got a few variables you're interested in. You need absolutely huge samples to account for the confounding variables like weather and has a fly got stuck on the windscreen and all this type of things which are inevitably going to affect the driver. You know, if it's raining if it's not. So the advantage of this type of controlled environment is that you can set up a really well controlled experiment and you're absolutely controlling all your variables so that the conditions are exactly the same between participants. Therefore, you can compare findings behaviour of different people precisely.
The highly controlled environment for testing is what makes IM@UCL’s simulator truly unique and incredibly valuable to anyone interested in driving related research. Stay tuned for upcoming episodes where you'll learn how the simulator could be used in areas outside of automotive, including maritime.
Thank you for listening to IM@UCL: The Podcast. If you would like to learn more about the research at IM@UCL you can check out their website at www.ucl-intelligent-mobility.com and/or subscribe wherever you are listening to this podcast so you can be notified when new episodes come out.
This episode was produced and hosted by myself, Cassidy Martin, with music from Blue Dot Sessions. It was brought to you by IM@UCL, which is part of UCL PEARL in Dagenham, and supported by UCL Minds – bringing together UCL knowledge, insights and expertise through events, digital content and activities that are open to everyone. A special thank you to Francesca and Jemima this month for sharing their time, knowledge, and insight. I hope you enjoyed listening to this podcast and feel like you learned something new. Like I have with everyone I've interviewed in the series.
Take care, and I'll see you again next month. Same time, same place.