brAIn blog: roundabouts v. traffic lights

Designing a smart city

Getting around town

During a recent walk around my neighbourhood, I thought about the things that can make moving around the city where I live, Cambridge, so irksome. Many pavements are in awful condition, and are partly obstructed by parked cars, skips or wheelie bins - not ideal for pedestrians, anyone with a pram or wheelchair users. Despite all the posters and signs telling us to keep two metres apart, it is impossible to achieve this while passing other pedestrians - I usually end up in the road. Every walk and run I go on is 'stop-start', as I have to stop at intersections to check for oncoming vehicles and cycles. Some cyclists turn into roads without signalling and at speed, which is terrifying if you're on foot. If I go out when the roads are quieter, it is likely to be dark, but there are some places where the street lighting is a bit poor and so safety becomes a concern. With respect to driving around the city, a typical journey involves a lengthy wait at at least one set of traffic lights, or involves a huge detour because of roadworks. Cambridge, despite having lots of cyclists, can be a scary place to go for a bike ride - cycle lanes, where they exist, end suddenly or are obstructed by vehicles, and only a few traffic lights have an extra phase in the signals permitting cyclists to move away first. Cambridge is a city that has evolved and grown over time, and so it has narrow roads and cobbled streets and insufficient space to enable everyone to safely travel around the city using their preferred mode of transport.

It is impossible to retrofit Cambridge with wide pavements and dedicated cycle lanes - there's simply no space for them. Similarly, the idea that we ban cars from the city is impractical. For now at least, we'll all just have to keep jostling for space.

But, what if you were designing a new town or city from scratch? How can you determine whether your city will be easy for people to move around in by foot, cycle or car? How do you optimise traffic flow while also ensuring everyone has easy access to services?

Making changes based on data

Recently, I learnt that roundabouts are generally safer than traffic lights. Apparently, U.S. crash data from 2017 to 2019 shows that 0.1% of crashes at roundabouts result in a death, while at intersections with traffic lights or stop signs, it is 0.4%. One reason for this is that at roundabouts, drivers tend to slow down and look around to determine whether it is safe to enter the roundabout, while at traffic lights, they often speed-up to avoid having to stop or rely only on the signals (and not their own senses) to determine what to do. Similarly, roundabouts are generally more environmentally friendly than traffic lights. This is because roundabouts allow vehicles to keep moving, i.e. they lead to smoother traffic flow. Traffic light signal timings are designed for maximum efficiency at the busiest times of day, i.e. the morning and evening rush-hours. During the rest of the day, drivers can be stuck at a red signal with the engine running, even though there's little traffic at the intersection, which means more pollution. It is apparently possible to move 50% more cars per hour through a roundabout than traffic lights, meaning that areas with roundabouts are less congested and quicker to get to (which means businesses in those areas are also more likely to be frequented).

As a result of this evidence, more new and growing cities in the US are implementing, or planning to implement, roundabouts at intersections instead of traffic lights. A large amount of data has helped city planners make a decision.

But not all data is good data...

In Cambridge, a bridge on a busy main road has been closed since June 2020 to private motor vehicles. The decision to close the bridge was apparently partly linked to the city's pandemic response: closing the bridge to cars, taxis and lorries would make it safer for people to get across the bridge by bike and on foot. Now, discussions are being held as to whether this closure should be made permanent. Some argue that the closure negatively impacts businesses on either side of the bridge, while others argue that the quieter roads means they're more likely to shop and eat on the road. Some say that side streets and other roads are busier because traffic is being diverted there, while others say that they enjoy the lower levels of pollution along the road. However, to me, it seems unfair to make a decision either way using data collected over the last nine months only, given how the pandemic has impacted 'normal' use of the road and 'normal' use of the businesses along the road.

Similarly, the UK's first 'Dutch-style' roundabout was opened in Cambridge in July 2020, in a bid to make a busy junction easier and safer for pedestrians and cyclists to navigate. Many are declaring it a great success, and there are already plans to build similar roundabouts elsewhere in the country. However, there has been one accident at the site since it opened. The levels of traffic at the junction are also much lower than in an 'ordinary' year, so it is difficult to really determine whether the roundabout is safer yet.

The US crash data comes from analysing accidents that occurred over many years and at many different sites. The decisions around the Cambridge bridge and roundabout are based on anecdotes and data collected at single sites over a much shorter period of time. But, what if you want to determine the impact of a new bit of infrastructure or a change to existing infrastructure? In this case, you want to determine the impact prior to starting the building work, so it's not possible to collect data beforehand.

Making decisions based on simulations

Simulations or mathematical models have been used for decades to, for example, plan, design and operate transportation systems. Such models can be used to determine, for example, the impact of a lane closure on a motorway, or to determine the optimum signal timing to avoid congestions.

My quick keyword search on a public database did not reveal huge volumes of European patents for simulation techniques (which use AI or ML) related to city and infrastructure planning or road traffic management. There could be a few reasons for this. One reason might be that some simulations may be very site or goal specific, and so they might not be generally applicable to other, similar sites or goals. For example, modelling the impact of a new traffic light system in a particular town may not be applicable to another town due to differences in town layout, existing infrastructure, population, number and type of road users, etc. Another reason may be that companies are not applying for patents (or getting patents granted) for these types of inventions in Europe.

However, something that many companies are doing is applying for patents in relation to autonomous vehicles, whether they are autonomous cars or grocery delivery vehicles, autonomous farm vehicles or even autonomous lorries/trucks.

Currently, simulations are being used to train autonomous vehicles. An example of this is described in European patent EP3345086B1. This patent describes a technique for simulating vehicle movement using a physical model and a machine learning, ML, method. When determining a next move for a next route or path segment of a next move cycle (e.g., a turn), based on the planned route information, a physical model is utilised by a data processing system to calculate or determine the vehicle's next status or state. The physical model refers to a set of rules or algorithms that have been configured to plan and control a movement of an autonomous vehicle based on a perception of the surrounding environment of the vehicle. The ML model is applied to at least a set of driving parameters associated with the planning and control data, as well as the driving conditions at the point in time (e.g., road condition, weather). The sideslip or skid under the driving scenario described by the planning and control data is predicted using the ML model. The planning and control data may then be modified based on the predicted sideslip to compensate such a sideslip. The vehicle is then controlled and driven based on the modified or compensated planning and control data.

Clearly, simulations are useful for predicting how systems (people, traffic, weather, etc.) may behave under different conditions. In some cases, the simulations can help people to design and build new buildings and cities. But there's now a desire to change existing systems - such as existing cities that may not have been designed well or old cities like Cambridge - so that they can operate more efficiently.

Making real-time decisions using AI

You've probably heard the term "smart city". These are areas which are fitted out with sensors monitoring all sorts of different things in real time (like traffic, pollution, parking space availability, etc.), and which use the sensor data to manage the operation of the city. The sensors are typically Internet of Things devices, which enables them to transmit the data they have collected in real or near real time. For example, traffic can be monitored by sensors in real time, and information on how long their journey will take using a particular route can be broadcast to drivers so that they can decide whether to continue along the route or take another one. Similarly, traffic data could be used to change the default traffic light signal timings to one that could speed-up traffic flow. Smart cities are touted as being able to improve energy distribution, optimise when refuse vehicles are deployed to collect our rubbish and recycling, and potentially improve air quality.

In order to make a decision based on the data collected by hundreds and thousands of sensors, you need to make sense of the data itself first. This is where AI tends to come in - artificial intelligence models can be used to analyse data, determine whether there are any problems, and identify possible solutions to those problems.

An example of how AI and sensors comes together can be found in European patent EP3144918B1, which relates to a computer system for monitoring a traffic system. The traffic control monitoring system can monitor the status of an entire traffic control system with an improved topological accuracy. The system makes use static visual sensors, such as cameras, to detect the state or status of traffic lights belonging to the traffic control system. An analytics module of the monitoring system uses ML algorithms to train a model for the signal phases of a traffic light. After the model has been trained, the module is able to predict signal states in the future. The system is able to send a message to a respective vehicle, where the message includes the current signal status and the at least one future signal status of at least one traffic light. The message is configured to influence the operation of the vehicle. For example, if the message indicates that the traffic light will switch from green to red in six seconds, the vehicle may automatically adjust their powertrain systems to the future situation. If a particular vehicle is able to reach the intersection under green light within allowed speed limits and traffic conditions, then the vehicle can adjust its speed to reach green light. If the vehicle is unlikely to reach the green light, the vehicle systems can notify the driver and prepare to slow down the vehicle speed and avoid crossing under red light while avoiding sudden and dangerous braking maneuvers. Taking advantage of the signal phase information, drivers are notified about the remaining time before the signal changes, increasing driver's awareness of an upcoming traffic situation and preparedness to react accordingly. In autonomous car applications the message might be used to automatically start and stop the motor at the optimal point in time. Such start/stop automation may save fuel/energy by affecting the vehicle to automatically slow down early enough by reducing energy supply instead of late braking.

Protecting your traffic and vehicle management systems

As shown above, companies are patenting their systems for managing traffic flow through cities, and controlling and training autonomous vehicles. There is lots of innovation going on in the 'smart city' space, so we can expect more patents in the near future which relate to using IoT data to make decisions that optimise cities. Similarly, many companies are using simulations to determine how objects or systems may behave before they are deployed in the real-world, and we expect more patents in this area.

Computer-implemented inventions like these are patentable in many jurisdictions, including in Europe. In Europe, in order to get a patent for a computer-implemented invention, such as using AI to control a vehicle or to simulate the impact of a new roundabout, we must show that the invention is a technical solution to a technical problem. I've talked about this requirement in a previous blog post. The European Patent Office has recently confirmed that the same requirement applies to both software that is used to control objects or systems in the real world, and to software that is used to simulate systems or object behaviour. In other words, it does not matter whether the software is purely a simulation or linked to something in the physical world. The key thing is that the purpose or effect of the software is technical. Thus, training an autonomous vehicle using a simulation may be considered technical, but using software to make financial predictions is very likely not technical (because business methods are not considered technical).

In the two example patents mentioned above, the technical effect of the method to train an autonomous vehicle may be that it enables the control a real object (vehicle), and the technical effect of the method to monitor traffic light systems may be that it enables a driver or autonomous vehicle to better react to changes in traffic situations and may also provide a greener way to operate the vehicle.

If you have developed any smart city or computer simulation technology and would like help patenting it, get in touch with me: parminder.lally@appleyardlees.com. At Appleyard Lees, we have a dedicated team of patent attorneys who specialise in protecting software and AI inventions, so give us a shout!