Roads with variable speed limits, designed to manage traffic flow, are normally adjusted according to simple rules, but a 27-kilometre section of the I-24 freeway near Nashville, Tennessee, is now overseen by an artificial intelligence.
Drivers on a busy US freeway have been controlled by an AI since March, as part of a study that has put a machine-learning system in charge of setting variable speed limits on the road. The impact on efficiency and driver safety is unclear, as researchers are still analysing the results.
Roads with variable speed limits, also known as smart motorways, are common in countries including the US, UK and Germany. Normally, rule-based systems monitor the number of vehicles on one of these roads and adjust speeds accordingly. One such road is a 27-kilometre section of the I-24 freeway near Nashville, Tennessee, which was experiencing a problem that besets many busy roads: when there are too many vehicles, phantom traffic jams appear when drivers brake, slowing vehicles to a crawl and risking crashes as fast-moving vehicles come up behind.
To address this, Daniel Work at Vanderbilt University in Nashville and his colleagues trained an AI on historical traffic data to monitor cameras and make decisions on speed limits, deploying it in the I-24 control room in February. Initially, the AI was tested in parallel with the existing software – not having control over limits, but telling operators what it would have decided – and faced teething problems.
“They look at it and for the first 5 minutes everybody’s like thumbs up, and then they start going to thumbs midway and then it’s thumbs down. So hey, it failed in the first 10 minutes,” says Work.
After some tweaks, the team launched a new system in March that has been able to operate unaided ever since. The AI works 98 per cent of the time, but will occasionally call for a change in speed limit that is larger than 10 miles per hour, in contravention of federal law.
“It’s a bad idea if the measured speed is going 80, 80, 80, 80, 20 – we don’t want that,” says Work. “We want it to go 70, 60, 50, 40, 30.” To ensure this, safeguards switch control back to the old system for the remaining 2 per cent of the time. “I can’t afford the phone call that comes and says: ‘It’s screwed up’,” says Work.
It isn’t clear how drivers have responded to the new system or whether it has improved traffic. The Tennessee Department of Transportation, which manages the I-24 , didn’t respond to a request for interview. Work says that data on the project won’t be released until later this year, as it is still being analysed, but he is positive about the results.
“I think that we’re just scratching the surface of a whole new way to operate freeways,” says Work. “It’s just absolutely transformational how these systems operate. Anything that we can do to reduce the number of crashes that happen on that roadway, the number of fatalities that happen on that roadway, is worth doing. It is in desperate need of attention, and that’s what drove the whole project from the first place: heavy congestion, a lot of growth and it is a corridor that sees a lot of crashes. Sitting by and doing nothing is not good for anybody.”
Oliver Carsten at the University of Leeds, UK, says that without more data on the trial results it is impossible to gauge whether the AI was a net benefit or detriment to safety, efficiency and reliability. But he says that some sort of variable speed limit system is key to maintaining safety on busy roads.
“There’s a well-known limit – 2000 vehicles per lane per hour – at which point you can suddenly go from everything running smoothly at 70 miles an hour to a total breakdown where everything comes almost instantaneously to a total stop,” says Carsten. “If you want to maintain throughput in very congested conditions, you need to bring maximum speed down and keep smooth operation of the road, because otherwise you get shockwaves. Essentially, just a few people slamming on their brakes will cause the traffic to grind to a complete halt.”
Reference