STRATEGIC THINKING

Prof Lee McCluskey on how the application of artificial intelligence can create the smartest method of managing traffic imaginable

In a number of sectors, such as transport and energy, there has been a lot of work in using sensors to accumulate, interpret and understand data – using artificial intelligence we can complement that by using the interpretation of that data in action. Instead of automatically interpreting the data, we do something with it – in the case of SimplifAI that’s control a system.

SimplifAI is a new form of urban mobility management system based on artificial intelligence and suitable for use as a smart city solution to improve reliability of transport networks and integrate autonomous vehicles into city transport. It deals in particular with non-recurrent congestion due to unforeseen or unplanned events. The main product is a real-time operational software tool that enables transport management operations team members to set an operational goal for the system and the tool will iteratively configure and reconfigure existing operational control assets to achieve that goal. The system uses existing real-time monitoring systems to track the progress towards achieving its goals and re-plans based on real-world monitoring of the impacts.

There’s an enormous amount of data around and there’s been an enormous amount of research and development undertaken that has harnessed that data to interpret, fuse and understand facts and information from that data. What we want to do is help authorities, those people that are stakeholders in the transport business, to manage more effectively their resources. We use the information and data as a starting point. For example, SimplifAI can automatically produce a strategy to avoid a predicted air pollution event in the middle of a busy city.

DEMANDING SITUATIONS
How SimplifAI puts its results into action is by taking a current state of the traffic network in terms of the data from sensors and signal controls. It uses historical data that will tell us the likely demands on the network – we call that our Initial State, but we then need to complement it with a desired outcome that needs to be achieved with a certain period of time. For example, if you have an incident on the network that has blocked a road we need to know the current state of the network at that point and also the expected demand on the network. The desired outcome in this scenario is that the traffic that is caught up in the incident is cleared and the expected traffic heading towards it is diverted or given alternative route options. SimplifAI creates and synthesises a strategy or strategies that can be enacted by the transport operators so that they can achieve their desired outcome.

The main advantage of SimplifAI is that it can do this in real time. We don’t need to spend a couple of days thinking about how to deal with this emerging incident – SimplifAI, given the Initial State and the desired outcomes, can work out a strategy within a few seconds. This strategy can then be transmitted to the network via the usual communications channels to the traffic signals.

3d rendering android robot thinking in office

TRIAL BASIS
SimplifAI is currently being trialled in four different types of scenarios and two of those are where we are assisting the transport operator with preplanned problems. As an example, in Manchester city centre you might have some planned roadworks during which there’s a football match between Manchester United and Manchester City and you would clearly want to create a strategy for the traffic signals to efficiently deal with the extra traffic that a match like that will doubtless generate. Staying with Manchester, in another example, Regents Road is a well-known and persistent bottleneck in the evening when traffic is heading out of the city centre towards the ring road and eventually the motorway to the West. SimplifAI comes up with strategies to alleviate that bottleneck. Perhaps more pertinent is how SimplifAI deals with unplanned incidents in real time which is, after all, what makes it unique.

No two incidents are the same – no two causes are the same, no two implications on the transport network are the same. SimplifAI creates strategies to alleviate the issues that accidents can cause, specifically those that block major roads. A huge fire is a good example. We would need to reprogramme all the traffic signals in the affected area – any incident that reduces the network’s capacity to handle traffic will mean that the SimplifAI engine has to create, in real time, a strategy to allow the traffic to flow as smoothly as possible.

Also, in the case of the aforementioned air pollution event, we need constraints in place that keeps the traffic below a certain level in the worst-affected area for the next hour or two to make sure the event doesn’t happen, or doesn’t worsen.

SPRINGING FORWARD
This is a very busy and crucial time for SimplifAI. The plan for between now and June is to bring all the pieces together. The central function, the automated planning function that relies on AI techniques, is just one of many pieces that need to be brought together to enable SimplifAI to leap from project to product. The next big step is to join the front end to the back end.

The accumulation of the information; the accumulation of the goals or desired outcomes; the requirements of the strategies are all parts of the front end. Assembling all the information that the automated planner needs is itself a major task. To be able to get accurate data that we need before the planning stage is a large part of SimplifAI – assuming we have all that, we come to the back end where all the plans have to be transferred into transport operator-speak, transferred into the language that can be eventually understood by the current technology which transfers individual plans for signal clusters at junctions to the on-street signals. That is not a trivial task. We’ve got to have constraints visible and explicit, both in the generated plan which respect the constraint of the signal strategies and individual plans for junctions actually on the streets. We need to demonstrate that the entire process works on real signals.

Prof Lee McCluskey has been researching and teaching artificial intelligence for many years – his PhD was in Machine Learning but over the last 20 years, and in particular the last 10, he has been concentrating on automated planning and engineering knowledge so that it can be used for automated planning. “Machine learning can be used in the engineering of knowledge so I’ve led several projects funded by various bodies – the European Union, InnovateUK and its previous incarnation as the Transport Strategy Board, all concerned with AI and automated planning. There’s a 20-strong AI automated planning group here at the University of Huddersfield, including 10 PhD students. My particular group that is focusing on SimplifAI are all publishing at leading conferences in AI so we have a long track record in AI and we bring that research expertise to the table.”

Prof Lee McCluskey is Professor of Software Technology at the School of Computing and Engineering, University of Huddersfield, UK and the academic lead for the SimplifAI project
t.l.mccluskey@hud.ac.uk

Author