Greater Manchester chooses Vivacity Labs to deploy AI-controlled road junctions to optimise traffic networks
- New world of travel sees more cyclists and pedestrians using junctions originally designed to prioritise cars and other vehicles
- Vivacity Labs’ AI to be used at ‘smart junctions’ to give priority to other road users, accommodating more foot/bike travellers
Vivacity Labs and Transport for Greater Manchester (TfGM) have announced the roll out of AI-controlled ‘smart’ traffic junctions to accommodate the increase of active travel modes, such as cycling and walking, in the city during the Covid-19 pandemic. The news comes after the initiative was declared winner of the Innovative Use of Technology award at the 2020 Intelligent Transport Systems, ITS (UK) Awards last week.
Using sensors with inbuilt artificial intelligence, Vivacity enables TfGM to anonymously identify different types of road users at selected junctions and control traffic signals to allow different modes of transportation to be prioritised as and when required. With more cyclists on the road as people avoid public transport, these ‘smart junctions’ will be able to give priority to people on foot or bike where and when appropriate.
Vivacity Labs' first-of-its-kind AI signal control system first went live early this year, before scaling to simultaneously control three, neighbouring junctions in the Blackfriars area of Salford in September 2020. This initiative also has the potential to reduce emissions and improve air quality in the Greater Manchester area. Congestion and queuing can be reduced by traffic signals that respond better and more quickly to changes in traffic conditions than existing systems.
This comes as part of a three-year Innovate UK co-funded programme (alongside Immense Simulations) to use AI to optimise traffic networks. Vivacity Labs has developed an algorithm that is able to adapt quickly to changing traffic conditions and efficiently implement high-level strategies at both local and city-wide scales.
There has been a nation-wide boom in the number of cyclists since the Covid-19 pandemic took hold. With people avoiding public transport when getting around towns and cities, and the daily commute changing for many in the longer term, innovation in transportation is a pressing need across the UK.
“Since the pandemic, commuter trends and traffic hotspots have changed completely, and cities need AI to help protect people no matter what mode of transport they take,” said Mark Nicholson, CEO of Vivacity Labs. “Our vision is to help cities implement critical policies addressing safety, air quality, sustainable travel, and congestion, at a hyper-local level.”
Digital Infrastructure Minister Matt Warman said: "Smart traffic technology is just one of the many ground-breaking areas the government is funding to pioneer new ways for artificial intelligence and 5G to transform our lives for the better. We're backing this initiative in (Greater) Manchester to improve the city's transport, reduce journey times and cut pollution. I look forward to seeing its positive impact and sharing the lessons across the UK."
Richard Dolphin, Highways Network Performance Manager at Transport for Greater Manchester commented: “Having already developed an innovative product in terms of their sensing technology, Vivacity has become well-versed in the relevant standards and have made impressive strides in the continuing development of their Smart Junctions system. We’ve been really impressed with how Vivacity has approached this, assessing current ways of working and addressing the complexities of managing a multi-modal transport network. Hopefully, this development will continue into something that will positively disrupt the industry and revolutionise active travel in urban areas.”
The project has secured additional investment to expand throughout the trial region via the Department for Digital, Culture, Media, and Sport’s ‘5G Create fund’ which was announced in July 2020. The project will scale up to an area of 20 junctions in Manchester by the end of 2021 and aims to demonstrate impact in the real world in the form of improved journey quality for all road users in this region.
Transport for Greater Manchester Press Office
0161 244 1055
Notes to editors
About Vivacity Labs
We’re revolutionising how transport is managed. Our award-winning Artificial Intelligence technology anonymously captures and classifies live transport usage, 24/7.
About Transport for Greater Manchester
Transport for Greater Manchester (TfGM) is the local government body responsible for delivering Greater Manchester’s transport strategy and commitments.
More than 5.6 million journeys are made across Greater Manchester’s transport network each day. It’s our job to do everything we can to keep the city-region moving and growing. We’re working hard to make travel easier through a better-connected Greater Manchester.
We've made transport simulation fast, accessible, and easy-to-use so you can make the best decisions for your fleet and infrastructure assets.
Media requests: For all media requests please contact Leah Jones: firstname.lastname@example.org
Intended for industry publications looking for more specific details on the technology
SCOOT & MOVA have dominated traffic signal control in the UK for the last few decades. Both have scenarios in which they work effectively, reducing congestion through coordination of multiple junctions (SCOOT) or through adaptively clearing queues and growing cycle times (MOVA).
However, many transport authorities are now trying to move beyond simply reducing congestion. Optimising air quality and prioritising active travel & public transport reliability are now critical to transport policies - yet MOVA and SCOOT struggle to do this effectively. Air quality optimisation with SCOOT has been trialled, but not rolled out at any scale. Bus priority in SCOOT, while well established, is a relatively blunt instrument, overriding optimisation for any other mode to provide late buses with green lights, and thus degrading overall system performance. Prioritisation for other key modes, such as cyclists, is not currently available. Meanwhile, it is well known that performance of SCOOT degrades over time, often by up to 30% - but recalibration is expensive and often not viable for many authorities.
At Vivacity, we are addressing all of these issues. Through AI, we have the potential to optimise for air quality, using our detailed datasets to understand emissions & set the AI to prioritise for this; or for cyclists / buses, providing weighted priority to reduce waiting times for these modes. Crucially, we can also optimise for congestion - giving transport authorities the choice to prioritise air quality, particular modes, or underlying congestion. Finally, AI can self-calibrate, maintaining performance indefinitely as the system self-improves and retrains following changes in the network.
In order to provide inputs to the machine learning algorithms, Vivacity have deployed our proprietary sensors across a large region of central Manchester. Each sensor provides anonymous real-time data feeds on counts, classifications, speeds, and journey times of vehicles, along with information on queue build-up. This wide range of data inputs are used to give the machine learning algorithms a broad understanding of the current situation, enabling some MOVA-like behaviour (short-term optimisations based on precise current vehicle positions, such as efficient stage closure) and some behaviour more similar to SCOOT, looking to optimise at a longer-term, regional level.
Reinforcement learning algorithms require training before they exhibit good performance. Before training, they would experiment randomly with the traffic signals, with predictably poor performance. To serve as a training environment, we have built a series of extremely accurate microsimulations, working closely with Immense Simulations. These simulations have been calibrated using months of our extensive & detailed datasets about every element of road movement, pushing the boundaries of microsimulation capabilities. We have now run over 8 million simulated hours to train these algorithms - the equivalent of observing and controlling a junction for a millennium!
Reinforcement learning (RL) as a field has been popularised by Google’s Deepmind, who have been using RL to demonstrate superhuman performance at games such as Go, which was previously thought to be computationally intractable, and the world’s experts typically play by intuition rather than logic. Deepmind’s algorithm is now the world champion at Go and has demonstrated genuine innovation with moves not previously known or understood by leading experts. We are using similar approaches to optimise traffic signals, training “agents” (a specific trained implementation of the algorithm) to control signals.
We have deployed on a junction in central Manchester and are gearing up to roll out across a series of nearby junctions. These junctions are currently under SCOOT control, last optimised 19/12/2016.
The agents are trained in a traffic simulation and their performance is currently measured against vehicle actuated (VA) control. We have seen a 22% reduction of waiting time for vehicles and pedestrians for a typical week based on observed demand levels and frequency at our first trial junction. The agents perform best during high demand, thus if performance is averaged across the full range of demands, the average reduction in waiting times (including pedestrians) is 33.8% when compared to VA.
Controlling three junctions in simulation has yielded similarly promising results. Three agents working side-by-side to control three neighbouring junctions have produced a 49% reduction in waiting times for vehicles compared to VA. This result is not unexpected since VA is not optimised for coordination between nearby junctions. For this reason, Vivacity are further developing their benchmarking suite to be able to robustly compare their system performance against industry standard algorithms which focus on coordinated control.
In the real world, initial analysis on the performance of the algorithm saw a reduction of average waiting times when compared to the performance of the existing system at these first trial junctions. As we continue to increase the hours of control, ongoing performance analysis is being conducted to assess the long-term impact and performance of the system.