At the end of last year we began work on an exciting new project, building on our smart mobility expertise: D-RISK. So today, I thought we would share some details about the project: why it’s important and what we will be bringing to the table.
A knowledge graph demonstrating a taxonomy of use cases (from project partner dRISK.ai LTD, 3 patents issued, 2 pending)
Last year, autonomous vehicles moved into Gartner’s “trough of disillusionment” in their hype cycle, as claims of full autonomy being imminent started to fall a little flat and the world realised we are still a little way off mass roll-out.
And what led to this recalibration of expectations? Well, probably several things, but one definite factor was the realisation of the scale of testing required to ensure that a self-driving car is equipped to deal with any situation it may encounter on a road. We don’t just mean the fairly predictable hazards or scenarios - cyclists, children running into the road etc. which could be seen to make up 99% of our “everyday” driving experience - but also the unexpected, rare-to-occur and complex circumstances; that 1% of driving experience.
The Edge Cases of Autonomous Driving
These unexpected or complex circumstances are often referred to as “edge cases” and although individually are very rare, there are potentially millions of them and together, they make up the majority of the risk in autonomous driving. As humans, we are fairly good at dealing with these, as we are able to use context to interpret situations or react according to set schemes. If you imagine a large animal walking out in front of you for instance, you might not know whether it’s a camel or a tapir, but you would know that you need to stop or go around it. An autonomous vehicle (especially one relying primarily on cameras to see the world), would first need to process a large number of dark pixels, realise that means there is an object ahead, then interpret that it’s a large animal and not say, a poster on the back of a bus, and only then make a decision on what to do. Once an autonomous car learns such a situation it would subsequently be able to react much faster (faster than humans in fact) but it will always be limited to the edge cases it has “learned”. So to truly move autonomy from hype to reality, we need to find a way to manage as wide of a range of edge cases as possible.
And that, in a nutshell, is what D-RISK is all about. The project is focused on creating a taxonomy of edge cases and critically, through the use of simulation, aims to then ensure that autonomous vehicles are able to safely respond and manage these situations.
Alongside our consortium partners dRISK.ai, Claytex, Imperial College London and Transport for London, we want to create the world’s largest driving scenario library, incorporating a huge amount of data from a variety of different data sources. These different sources of information will be collated and structured before being fed into a comprehensive knowledge graph of all autonomous vehicle risk scenarios.
From that knowledge graph, which might look something like the one above, representative test cases can be fed into one of several test environments, both physical and virtual, to directly test the vehicle control system.
The end goal of all of this work is to maximise the safety of autonomous vehicles on our roads by not only identifying edge cases, but understanding what is most important to the public. Or to put it another way, we’re creating the ultimate driving test for self-driving cars.
Bringing Humans into the mix
So where does DG Cities fit into all of this? Well, as ever, we strongly believe that technology needs to be developed and deployed with a consideration of how it fits into the community it aims to serve. This is incredibly important when it comes to autonomous vehicles, because without people trusting in them being safe and equipped to manage our roads, their widespread deployment will always be limited.
That’s why we were so keen to introduce crowdsourced human-data inputs into the creation of the D-RISK taxonomy, alongside the many other different sources of data sources and sophisticated research being undertaken across the project. Our role therefore will be undertaking critical community engagement work, going out and talking to local citizens at focus groups and interviewing road users. Asking them what situations they encounter when driving or crossing junctions, what scenarios they are worried about when it comes to self-driving cars and perhaps the more unusual things they have witnessed on the roads. Don’t forget, in this instance we’re not looking for the “oh you always see this” scenarios but the “you’ll never believe what I once saw” scenarios. So we believe that humans and our capacity for remembering and conveying stories can provide varied and colourful inputs, complementing the many other data points the project will be collating.
These human-data inputs will contextualise and validate other data sources, contributing not only to the scenario library we are building, but critically, can then be used to “train” autonomous vehicles. Through ongoing community engagement methods, we will be keeping citizens up to date on how we’re putting autonomous vehicles through their paces, ensuring they can handle the scenarios they have flagged to us. In this way, human-inputs will be used to design the ultimate CAV service delivered to our roads. We believe it’s the first time such an approach has been taken and are excited about not only the richness of data we will source but the strides this should enable us to take in building trust between human and machine.
D-RISK is set to be an exciting project and we’re proud to be working with another set of best-in-class partners, to be delivering work that will significantly advance the realisation of full autonomy on our roads.
If you’re embarking on your own CAV trial or would like to understand how to integrate autonomous vehicles into your city, we’d love to hear from you. Drop us a line at info@dgcities.com.
