by Australian Integrated Multimodal EcoSystem (AIMES), University of Melbourne
We’d all like to spend less time sitting in traffic, and now a new Artificial intelligence (AI) application offers hope for congestion-weary commuters
Peak hour driving, especially in Sydney and Melbourne, can set even the calmest among us on edge. It’s no wonder, with freeway speeds regularly slowing to around 30km per hour and journeys taking around 50 per cent longer than they would in freeflowing traffic.
Governments are pumping billions into much-needed infrastructure upgrades, but these improvements will take years. Fortunately, new ways to speed up the roads we already have, right now, are also emerging.
AI and machine learning analysing traffic patterns
A particularly promising tool is a new AI application developed in Melbourne that can predict traffic up to three hours in advance. It provides detailed information customised to any route, and could help future commuters avoid hotspots, and governments ease traffic jams before they happen.
Unlike commonly used tools like Google Maps, the new application uses machine learning and artificial intelligence to understand traffic patterns. It learns and adapts as it goes to ensure its predictions are as accurate as possible.
The new tool is the result of a collaboration between the University of Melbourne’s Australian Integrated Multimodal Ecosystem (AIMES), the Victorian Department of Transport, Telstra and Melbourne-based AI firm, PeakHour Urba Technologies, who developed the core AI engine and its predictive capabilities.
AIMES brought together the project partners. Professor Majid Sarvi, from AIMES, said that most of the time tools like Google Maps get it wrong because they use historical data, which is often inaccurate and vague.
“There’s such a build up with traffic congestion, by the time you’re on the road the travel time is wrong,” Professor Sarvi said.
Five years of rich data
The difference with this tool is that its predictions change dynamically based on what’s happening on the road, making it reliable across a whole transport network. “It uses machine learning in a way that we couldn’t have dreamed of, even two years ago,” Professor Sarvi said.
“But this system is self-learning and its predictions are beautifully accurate.” In a city like Melbourne that’s no small feat, with its road network comprising around 20,000 unique links like junctions, roundabouts, slip roads and more.
Its development relied on real-time data gathered through AIMES’s ‘testbed site’ – a network of over 2,000 data sensors located across 100km worth of inner-Melbourne streets.
The site has been operational for five years now. The data gathered at this site is among the richest in the world, and it provided developers with the kind of information they needed in order to understand how traffic behaves under a wide range of different scenarios.
And it doesn’t just apply to inner Melbourne. “The data can be extrapolated out to other areas. It shows the possibilities when you have access to the right data,” Professor Sarvi said.
Global interest in the technology
Indeed, while the tool is up and running in Melbourne now (although not yet publicly available), it only takes five days to learn traffic patterns in other cities. The team is already fielding interest from overseas.
This adaptability is thanks, in part, to being hosted in the cloud and able to operate using minimal data. The application uses whatever data is available to it, with Amazon Web Services providing the scalability to ingest, store, and process
large amounts of traffic data.
With the quality of mobile-derived traffic information growing fast, the developers are anticipating even more data will be available in the next few years. Smartphones, connected cars, vehicle-to-everything and vehicle-toinfrastructure technologies, are likely to account for an increasing proportion of its real-time data in years to come.
But even using current data points, the new tool offers hope for congestion-weary commuters. The university’s researchers have found it could improve efficiency within the transport network by 20 per cent.
“Shaving one fifth off travel time would be welcome news for commuters, but it also could improve the bottom line for freight operators and fleet managers,” Professor Sarvi said, suggesting the new tool could help them optimise their networks and increase efficiencies.
Helping cities make traffic decisions
Perhaps the most enthusiastic users will be governments, who will be able to use the tool to improve how they manage congestion.
Currently, most cities rely on humans monitoring traffic networks and making real-time decisions about traffic flow to try and control congestion, for example by changing traffic light sequences.
“The idea of people sitting in a control room, watching live video feeds and making decisions to keep traffic flowing, is just not sustainable,” Professor Sarvi said.
Instead, he said the new tool’s prediction capabilities will mean cities can proactively manage traffic, rather than only responding once congestion has already started to build up.
“It highlights the hotspots so the traffic can be cleared in those corridors and everyone gets home faster. The next steps will be handing end-to-end traffic management over to AI.
“That’s the holy grail. I would imagine very soon AI can alsooptimise the traffic signal for us and provide full end-to-end prediction management. AI is simply much better at it than us humans,” Professor Sarvi said.
With the technology working so well, the team sees their biggest challenge in getting it adopted for real-world use. They have been working closely with the Victorian Government and are hoping to see it in action in Melbourne soon.
“We could get this up-and-running in Melbourne as a world-first; if we don’t, some other country will beat us to it,” Professor Sarvi said. “We’re really hoping to see it here first.”