�����

We could increase traffic efficiency by up to 45 per cent, Yang Bo says, discussing a futuristic lightless intersection he modeled at A*STAR’s Institute of High Performance Computing (IHPC) last year. Creating smoother traffic flows could also save millions of liters of fuel that would otherwise be consumed globally by vehicles idling at traffic lights each day. Indeed, so could the ideas of multiple teams at the IHPC working on computational modeling to improve road use, which include several looking at the ‘hive brains’ of taxis and ride-sharing vehicles.

The IHPC is fortunate, �����is a major test bed for intelligent transport systems. Its roads pulse with information and subtle adjustments: smart intersections vary their cycles according to the flow of traffic and intelligent ramp meters record each car, while a congestion tax, politically contentious in other cities, is collected by electronic gantries daily. This is all managed by the Land Transport Authority (LTA) — who have 164 kilometers of expressways and road tunnel systems wired for data collection and video surveillance. The information is fed to operators who smooth flows and send assistance to motorists in trouble. Because of this coordinated effort, �����— one of the densest cities in the world — sits at a comfortable number 55 in TomTom’s congestion world ranking, well behind less populous cities such as Sydney, San Francisco and Auckland.

Eliminating traffic lights from intersections

Yang proposed his lightless intersection in 2017, along with a colleague, Christopher Monterola¹. Cars continuously move through these junctions, and while human drivers still steer, acceleration, deceleration and interactions with other cars are controlled by a beacon installed on the dash and at the intersection. He says the efficiency gains are found in reducing redundant waiting for light changes and human factors such as “phantom traffic”, in which drivers slow unnecessarily at the fringes of a congestion, with rippled effects slowing cars upstream.

It’s a response to a hopeful, but uncertain future as autonomous vehicles begin to penetrate the market, Yang says. “The question is: How do we make a new system that has the least required modification to vehicles until we fully implement driverless cars?” While tech behemoths such as Didi and Google race to develop systems to guide driverless cars through such quandaries, Yang says the aim of his model is to make something practical today and for the anticipated “long transition period” to fully driverless technology. “If you have a high technological barrier, say you have a system that needs an autonomous vehicle, then it’s going to take a long time to implement, because there will be cars that aren’t driverless for some time,” he points out.

In 2016, the world’s first driverless car policy emerged from the United States and outlined five levels of autonomous vehicle ranging from regular cars with driver assistance (Level 1) to fully automated machines with no steering wheel (Level 5). At its simplest, Yang’s system only requires vehicles to be at Level 1 — the intelligent car beacon takes over some aspects only in the intersection ‘zone’ and protects passengers using an algorithm that creates a mathematical repulsion between vehicles.

Because the zone is highly localized, this system is also easy to implement and requires minimal energy to run. Yang adds that his modeling of the intersection sees overall efficiently gains even if today’s cars (he calls them ‘legacy cars’) make up 70 per cent of the traffic. He explains that for today’s vehicles, these intersections would incorporate something like a traffic stop system. The caveat is “it will be a little slower for older cars, but that will encourage people to upgrade.”

Without real-world implementation and accounting for vehicle-pedestrian interaction, Yang’s intersection is still theoretical. But in Singapore, ideas like this can become reality; the city’s advanced thinking on the road means it has bucked dominant world trends and actually improved its congestion levels in recent years.

The red carpet is also being laid out for companies diving into the autonomous vehicles market by the �����Autonomous Vehicle Initiative (SAVI), a joint partnership between the LTA and A*STAR. They see �����as the logical home for the world’s driverless statutory and technical firsts, which indeed it may well be. At the moment, there are only a handful of official city-based driverless trials in the world; this year �����will host both the world’s first driverless taxi trials for the MIT spin-off nuTonomy and commercial driverless cars tests for electronics and software company, Delphi Technology.

Optimising taxi roaming

This forward-thinking approach to its roads is typical. For decades now, Singapore’s government has been extraordinarily clever with its road data. For example, each of the city’s more than 20,000 taxis are required to transmit its GPS location and working status every 30 seconds to the LTA, so that it can monitor a key mode of transport in a city where fewer than one in 10 own a private car. Some of this data is being harnessed by the IHPC to come up with ways to improve taxi occupancy.

In 2016, a team led by Qin Zheng, a senior scientist also at the IHPC, started by looking at a week’s worth of taxi data. To examine the 3.6 million-odd data points, they worked with the Fujitsu-SMU Urban Computing and Engineering Corporate Lab in a collaboration between A*STAR, technology giant Fujitsu Limited (Fujitsu) and the �����Management University (SMU). Zheng’s team used the taxi data to train a learning neural network they devised — called Fusion Architecture for Learning and Cognition with Alternative MemorY (FALCON.Amy) — to predict pick-up hotspots for taxi drivers.

At peak demand, street pickups become much more efficient than bookings systems, which include disruptors such as Uber and Asian counterparts, Grab and Didi. IHPC and other studies have shown that in high-demand conditions, vehicles will often pass a number of potential passengers to get to their booked passenger.

The collaboration soon spawned an app using FALCON.Amy, called the Driver Guidance System (DGS). Within the year, the app was already directing taxis to roads where they’re likely to find a fare, and being used in the National Taxi Association (NTA) SkillsFuture Training Programme.

Zheng says Singapore’s taxi association is big supporter of the project. “They are very anxious to improve technology for their drivers,” he says. While alleviating congestion by decreasing unnecessary road use, reducing empty taxi roaming could also boost profitability for taxis, which are struggling in an era when disruptive technologies are cutting into their bottom line.

The DGS was primed to predict fare locations using roughly two years’ worth of taxi data. Numbers from a free trial for taxi drivers showed significant improvements for cab drivers during off-peak hours; those using the app decreased their average empty roaming time in the city between midnight and 6am from 17 to 12 minutes. Early feedback, collected through discussions with drivers who participated in the trial, was also predominantly positive, particularly for new taxi drivers who don’t already know where to find fares. However, some taxi drivers reported a reluctance to miss out on booking fees.

This doesn’t bother Zheng. While, it seems that companies like Uber and Grab are here to stay for the mid-term, he says the long-term thinking is that they will be used for another five to ten years, after which the city will move to driverless taxis. “Imagine there are no drivers; then all we want is to optimize the social benefit,” he says. In that future scenario, systems, including the one Zheng’s team trained for this project, will be less constrained by individual drivers’ needs and could form a part of very efficient future traffic systems run largely by advanced algorithms and artificial intelligence.

However, in countries such as Fujitsu’s country of origin, Japan, the data they need to train a system like Zheng’s will have to be drawn from a wider variety of stakeholders than in Singapore, including a number of different taxi or transport companies. But, while �����has one of the world’s most centralized traffic data sources, the base information needed to run Zheng’s program exists in any city with GPS-linked taxi-like services. “If we can get hold of Uber or Grab data we can apply it,” he says.

Taxi numbers reduced by more than half?

Yang adds that increased sharing of vehicles has the potential to further winnow traffic. One 2015 study showed that in New York, assuming people were willing to share taxis whenever possible, four-seater taxi demand could be reduced to just 15 per cent of the current fleet.

In Singapore, says Yang, if a ride-sharing algorithm he and his colleagues have developed² were to be adopted by 50 per cent of taxis — in a system in which more than one person is an individual paying passenger — the 15,000 taxis on the road at any one time could be reduced to 6,000.

But he’s quick to explain that this number needs to be taken with a grain of salt: “We’re still in the process of looking at lots of information, particularly traffic conditions and commuter boarding and alighting behaviours.” That’s their next step, he says “adding more and more information to make the simulation more accurate.” Yang also points out, “the tricky thing about �����it is that it’s a relatively rich society, so people don’t really want to give up their privacy to save a few dollars.”

But Yang’s work shows that ride-sharing taxis may find a foothold during peak hours and bad weather, when passengers are faced with surges in demand. His model suggests that during busy periods, gains in shorter wait times could far outweigh the time cost of accommodating another taxi passenger, not to mention the real cost savings.

Taking all of those taxis off the road will also reduce traffic and travel times, says Yang. But those benefits will only become evident if advanced modeling like the IHPC’s “can convince policy-makers to cultivate a culture of ride sharing”, he says. “Our results show that a good algorithm for taxi ride-sharing can really help,” he explains, “because increased ride-sharing is not only for the common good but can also be immediately helpful to individuals as well.”

A culture of hyper-organized solutions has always been Singapore’s strategy for dealing with limited space. As the world approaches a watershed for driving technology, a dense and still fast-growing population means that �����is incentivised to remain the world’s most cutting-edge place for intelligent traffic. The government also continues to invest heavily in advanced technology: in 2016, for example, the LTA announced a new S$6 million (US3 million) project to install units into all Singaporean cars, so a satellite tracking system can determine congestion fees and automatically deduct charges for curb-side parking. Driverless cars are perhaps still a decade away, but �����and its researchers are positioning themselves as front runners in the road race towards the smartest cities of the future.