London’s road network is one of the oldest in the world, and for the most part was designed and laid out when the city’s population was much smaller than it is today, and in a vastly different built environment.
London’s ancient origins mean that it doesn’t benefit from the grid layout common throughout the US. London roads — in common with much of the rest of Europe — are the result of medieval foundations, and have been forced to continually evolve over the centuries as the city itself grew.
This busy labyrinth that is today’s London results in complex interactions between traffic and other road users as they attempt to navigate through the maelstrom of one of the western world’s most densely populated cities.
This complexity presents significant challenges for the development of autonomous vehicles (AVs). But succeeding in this challenge is fundamental to creating safe autonomous vehicles which will redefine how we travel around London, but also thousands of other towns and cities across the world.
One of the main issues that AVs will have to tackle in London is the sheer number of different types of road user and associated behaviours within each scene they encounter. From cars, double decker buses and pedestrians, to wheelchair users, road workers, animals and emergency service vehicles, the number of scenarios AVs will have to deal with are virtually limitless.
Take cyclists, for example; this one category of road user can be split into a series of different subcategories, each with their own distinct behaviours. A courier cyclist will frequently behave very differently to a company executive riding to a meeting on a rented public bike. And with cycling now the leading mode of transport in London’s rush hour, it is key that AVs understand this user group and learn how to share the road with them safely.
Other factors that must be taken into consideration when developing AVs to take on London’s roads include:
- Social cues and interactions
Human road users (in any capacity) rely a lot on social cues when communicating with and reacting to other road users, quite often subconsciously. These might be as subtle as eye contact or a small nod of the head, but such micro gestures are part of the fabric of how our road networks function. So AVs need to be able to read these cues too to enable optimal safety.
- Multiple simultaneous manoeuvres
Unlike in quieter suburban settings, vehicles in dense urban environments must be able to make multiple road manoeuvres at the same time. This might take the form of passing a parked vehicle whilst slowing for roadworks ahead and navigating a sharp corner, for example.
- Continuously changing road maps
Continuously updating detailed maps to account for the myriad of roadworks, road closures and diversions that happen across London every day is a huge challenge. Therefore, our AVs cannot rely entirely on prior mapping to navigate our cities, but must be able to read and understand the road situation in front of them at all times to take account of any changes.
- The glorious British weather
Our varied and often challenging climate is inescapable. AVs will need to be able to cope with a range of conditions, from rain and fog to snow and bright sunshine. The nuances of different weather conditions, such as reflective or very deep puddles, or sun shining through early morning mist, must all be taken into account to ensure the safe operation of AVs.
- Hyper-localised behaviours
The more complex the urban environment, the more likely you are to have hyper-localised behaviour that varies from one part of a city to the next. Road users in North London may be seen to behave differently from those in South London (for example), and so local knowledge from learned behaviour of road users is crucial to the development of AV technology.
We’re building AV technology that will avoid causing incidents wherever it’s in use, but there’s more to safety than not driving in to things. Other road users have expectations for how our vehicles will act and react, and driving too slowly or too hesitantly can itself cause problems, making other road users more likely to take risks.
Key to safety is being able to have confidence in our understanding of how complex environments will develop. In order to do this, our vehicles will constantly observe and learn about specific behaviours pertinent to the locations that they will be driving, e.g. the way that pedestrians tend to act around schools in the morning or pubs and clubs in the evenings.
Engineering our AVs to not only understand, respond to and navigate London’s roads safely is no mean feat, but it’s one we must solve — with safety as an absolute priority, always. Arguably there is no greater test for our technology; but if it works in London, it can be made to work anywhere in Europe.
- Ben Peters, VP Product, FiveAI