×
About Motorway
Five kilometres (about three miles) of British motorway, with one, two or three lanes and traffic keeping left. JOINING has a slip road coming on from the nearside into lane 1. APPROACHING AN EXIT has a slip road leaving from lane 1, with the advance signs at 1 mile and ½ mile and the countdown markers at 300, 200 and 100 yards. ROADWORKS closes the offside lane at a line of cones. On the first two, HOGGERS puts some drivers in lane 2 for good. The wide camera shows the whole stretch; the junction camera shows the few hundred metres around the slip road, where you can watch each lane change; the pen recorder colours the road by the speed of the traffic on it, for all lanes together or one lane at a time.
WHAT THE DRIVERS DO
Every driver follows the car in front by the Intelligent Driver Model (Martin Treiber, Ansgar Hennecke and Dirk Helbing, 2000), the same model as in Phantom Jam , with a small random wobble. Plain IDM brakes far too hard whenever somebody pulls in close in front, which on a road with lane changes would make almost every change a "hard brake", so the drivers here use its refinement, the ACC model (Arne Kesting, Martin Treiber and Dirk Helbing, 2010): a driver who can see the car that pulled in isn't braking eases off instead of standing on the brakes. Cars want 70 mph, the UK limit, give or take about 5 mph (60 to 82 mph); they keep a 1.2 second time gap unless you change it. One vehicle in ten is a lorry, 16.5 m long, keeping a time gap 0.3 s longer than the cars. Goods vehicles over 3.5 tonnes must have a speed limiter set at 56 mph (90 km/h; when limiters first came in, only lorries over 12 tonnes needed one), so the lorries here never go faster than 56 mph, even though the legal motorway limit for goods vehicles over 7.5 tonnes in England and Wales is 60 mph.
Each driver decides when to change lane by MOBIL ("minimizing overall braking induced by lane changes", Arne Kesting, Martin Treiber and Dirk Helbing, 2007). A driver changes lane when the move would let it accelerate more than staying put, after taking off a share (its politeness : 0.2 here, the page's own choice; the paper explores values from 0 to 1) of whatever the move costs other drivers, and only if nobody would have to brake harder than 4 m/s². The paper has a version for countries that drive on the right with "keep right unless overtaking"; this page uses it mirrored for the UK:
Keep left (Highway Code rule 264): every driver has a small standing wish (0.3 m/s²) to be one lane further left, so an empty motorway drifts into lane 1 and drivers return after overtaking.
The lane further out has priority : moving out, only the loss to the driver behind in the new lane counts; moving in, only the loss to the driver left behind in the old lane.
No overtaking on the left (rules 267 and 268): nobody draws level with, or pulls ahead of, a slower car in the lane to its right. Rule 268 relaxes this in congestion, when the lanes are moving at similar speeds and you may keep up with your own lane. The model approximates that with a fixed switch from the paper: once the lane to the right is below 60 km/h (about 37 mph) the ban lifts. The paper enforces the ban by making the driver follow the car in the next lane with a full time gap; here the driver only has to avoid drawing level, which is closer to real driving.
In queues , as in the paper, the keep-left rules are switched off: when a driver and the lane it would move into are both below that 60 km/h, both drivers behind count and there is no keep-left wish.
Lorries never use lane 3 of a three-lane motorway (rule 265).
A lane change takes the car across in about 1½ seconds on screen; in the sums it belongs to its new lane at once, as in MOBIL. No driver changes lane more often than once every 3 seconds, and a change must gain at least 0.2 m/s² (the paper suggests 0.1; with the wobble that had drivers changing lane every few hundred metres).
JOINING
Joining drivers come down the slip road at about 50 mph, speed up, and move into lane 1 anywhere along a 250 m merge section when MOBIL says it is safe; if they run out of slip road they stop at its end and wait for a gap (rule 259: give priority to traffic already on the motorway, match your speed). Half the drivers in lane 1 (every other car, so runs repeat exactly) ease off to let a joiner in front of them; the others don't. When more cars arrive than the road past the junction can take, a queue grows back up lane 1 from the slip road, and drivers move out of lane 1 to get past it. With one lane the model breaks down harder than real roads do (joiners race down the slip road past a crawling queue and cut in at walking pace); UK motorway main carriageways nearly always have at least two lanes.
APPROACHING AN EXIT
A share of the drivers (TAKING THE EXIT) want to leave at the junction. Rule 273 says to move into the left-hand lane well before your exit. MOVING OVER sets where they start insisting on lane 1, from EARLY (1 mile, at the first sign) to LATE (200 yards, the second countdown marker). The signs and the countdown markers measure to the point where the exit slip road leaves lane 1. Before its point an exiting driver drives like everyone else; after it, it only moves left, wants to badly enough to accept a slower lane, and slows to a stop at the start of the slip road rather than go past, waiting for a gap. Drivers who move over late are also less polite about it: from EARLY to LATE the politeness falls from 0.5 to 0 and the braking they will force on the driver behind rises from 2 to 5 m/s². This is the page's own way of describing "late"; real drivers vary. A driver still two lanes out with 100 m to go, or who has waited half a minute, gives up and carries on to the next junction (MISSED EXIT). Exiting drivers in lane 1 slow to 60 mph in the last 200 m and to 40 mph on the slip road, which has no queue at its far end. As in JOINING, half the drivers let an indicating driver in.
An exit run lasts 15, 30 or 60 minutes of road time (LENGTH on RUNS SO FAR; 30 to start with); the first 5 fill the road. Measured from minute 5 to the end: the through traffic's average speed in each lane over the 1½ km before the exit, stops (a car dropping below about 1 km/h) and hard brakes (slowing at more than 2 m/s²) in each lane over the whole stretch, the flow past the exit and the drivers who missed it. Cars arrive on schedule whatever the road is doing; if it is full at the start they wait there, as they would in a queue further back. Time lost is measured on a fixed group: every driver due at the start of the stretch from minute 5 until 6 minutes before the end (minutes 5 to 9, 5 to 24 or 5 to 54), each timed from when it was due (so any wait to get on counts) until it reaches the exit, minus the time that would take at its own wanted speed. A driver in that group still short of the exit when the run ends counts with the time it has lost so far, so the average is then a minimum: RUNS SO FAR marks it ≥. When a run ends, KEEP GOING lets you watch the same road carry on; nothing after the end goes into RUNS SO FAR.
One run can be lucky or unlucky, so pressing RUN also measures the same setting, for the same length, on ten traffic patterns (the same ten arriving streams of cars for every setting) in the background, and RUNS SO FAR shows their average, with the smallest and largest time lost. An hour-long setting takes a minute or two to measure. PAUSE pauses this measuring too, and RESUME carries it on from the pattern it had reached; RESET stops it and clears its unfinished column. "Stop-and-go" counts the patterns in which lanes 1 and 2 had 100 or more stops per 10 measured minutes. The cameras show one pattern at a time; NEW TRAFFIC steps through all ten.
WHAT THIS MODEL SHOWS, AND WHAT IT DOESN'T
In this model, at the starting settings (4,500 cars an hour on three lanes, a quarter of them leaving, 30 minutes), across the ten traffic patterns: with everyone moving over EARLY, through drivers lost about 23 s each getting to the exit (22 to 23 s across the patterns), leaving drivers about 31 s, nobody missed the exit and the road never tipped into stop-and-go. With everyone moving over LATE, through drivers lost about 68 s each on average (from about 25 s to about 2 minutes) and leaving drivers about 75 s; the cut-ins tipped lanes 1 and 2 into stop-and-go in 8 of the 10 patterns, and there the through drivers' average rose by roughly 15 s to over 1½ minutes against EARLY. In the 2 that coped it rose by a few seconds to about 15 s. LATE also brought about 4 times the hard braking in lanes 1 and 2 and about 65 missed exits per run. Length matters because the jam, once started, keeps growing: over 15 minutes LATE tipped in 5 of 10 (through drivers lost 37 s on average, EARLY 22 s); over an hour it tipped in all 10, the queue reached back 1 to 2½ miles (in several of the 10 all the way to the start of the stretch) and through drivers lost at least 3 minutes each on average, while EARLY stayed at 23 s. Which patterns tip is sensitive to small details: even the last digit of a browser's arithmetic can tip a pattern that is on the edge, so your own figures may differ a little from these. That LATE tips more the longer it runs, and EARLY doesn't tip, is the steady result. Moving over EARLY puts everyone leaving into lane 1 well before the exit: with a large share leaving, lane 1 then queues back from the exit while lanes 2 and 3 run freely.
That is not the same question as merging at roadworks (the ROADWORKS tab, below), where a lane really does close. At an exit nothing closes: the late mover leaves a lane that carries on, and cuts into lane 1 near where it is busiest. The model is a picture of the mechanism, not a forecast for any real junction.
MIDDLE-LANE HOGGERS
HOGGERS (on JOINING and APPROACHING AN EXIT) makes 5, 10 or 20% of the car drivers carrying on past the junction sit in lane 2. Their keep-left wish is replaced by a strong wish for lane 2 (1.5 m/s², five times the keep-left nudge; the page's own figure): they leave it only when the car ahead holds them up a lot, and go back to it as soon as there is room. Their wanted speeds are drawn exactly like everyone else's, so what they change is their lane and nothing else. Lorries and drivers leaving at the exit never hog. Who hogs comes from a random stream of its own, so every other car is the same car with or without hoggers, and the 10% hoggers are the 5% plus some more. Highway Code rule 264 says to keep in the left lane unless overtaking and to return to it when it is safe. Hogging can amount to careless driving: when the police in Great Britain were given a £100 and 3-point fixed penalty for careless driving in 2013, the government named middle-lane hogging as an example; the more serious cases still go to court.
Drivers in lane 1 may not pass a hogger on its left (rules 267 and 268; rule 268 makes an exception in congested traffic, where neighbouring lanes move at similar speeds and you may keep up with your own lane even if that means passing traffic on your right, and the model does the same once the lanes are crawling). So a driver held back by a hogger beside it counts what it could do two lanes out: it moves out behind the hogger, then to lane 3, passes, and comes back in. With two lanes there is no way round. UNDERTAKE lets drivers pass hoggers on the left at any speed; the Highway Code says not to, and the page counts no risk for it, only time. When you RUN an exit setting with hoggers, RUNS SO FAR also measures the same traffic with none, and compares the same drivers in the two runs, the hoggers and everyone else separately. LANE USE gives each lane's share of the miles driven before the exit.
In this model, comparing the same drivers over the ten patterns: at the starting exit settings (4,500 cars an hour, a quarter leaving, moving over EARLY, 30 minutes) 20% hoggers made little difference to everyone else: drivers who didn't hog took 0.6 s longer each (the hoggers themselves 4.5 s), and lane 1's share of the miles fell from 35% to 33%, because in traffic that heavy everyone else fills lane 1 anyway. Nearer the road's limit (5,700 an hour, nobody leaving) 20% hoggers cost the drivers who didn't hog 18.8 s each and the hoggers themselves 24.4 s (10% hoggers: 7.8 s and 11.7 s), and lane 1 carried 28% of the miles instead of 32%. With UNDERTAKE the cost to everyone else disappeared (they were 0.8 s quicker), but the hoggers still lost 4.1 s each, held up by slower traffic in lane 2 that they would rather follow than pass. What hoggers cost here is the drivers held up behind them and the extra lane changes to get round, not a lane closed to traffic. In these runs hoggers spent about 9 in 10 of their seconds in lane 2.
ROADWORKS
The offside lane (lane 2 of 2, or lane 3 of 3; LANES can't be 1 here) closes at a taper of cones 150 m long, 4 km along the stretch, and stays closed to the end. Lane-closed signs stand at 800, 400 and 200 yards, and a temporary 50 mph limit starts 300 m before the cones. 50 mph has long been the usual limit through narrowed lanes at UK motorway roadworks; National Highways now aims to set the "highest safe speed" for each set of works, which can be higher. Past the cones the lane is narrow, with workers beside it, so drivers build up a time gap 0.6 s longer than usual over the first 350 m (the page's own figure; the results below also try 0.3 s). That sets how many cars an hour get through: for two lanes down to one, about 1,380 to 1,390 an hour in this model with the 0.6 s gap, whatever the merging method, and up to about 1,570 with 0.3 s.
Every method uses the same drivers and the same cooperation. At the cones (from 200 m before where a closing-lane driver has to move over) each closing-lane driver is paired with the open-lane driver that will follow it in: the first one behind it that hasn't already let someone in, taken in order from the front, so the drivers go one from each lane. The partner eases back early and smoothly to leave room for its pair plus a short gap (1.5 m and 0.3 s, as drivers leave when letting someone in at low speed), braking by at most 3 m/s², and drops back to its normal gap once the pair is in front. The closing-lane driver keeps just behind the open-lane driver ahead of the space, at that driver's speed, and moves across on the move as soon as it is safe; with nobody within 150 m behind in the open lane it just moves over. Further back , a driver beside the closing lane makes room for the closing-lane driver that will move into the space in front of it once that driver is within 100 m of where it must move over, keeping its normal following gap behind it; the driver it lets in has had its turn for the next 100 m, and a merging driver drops back rather than squeeze in front of it unless there is a gap of 30 m or more anyway. Every merging driver will make the driver behind brake by at most 3 m/s², at any speed.
IN TURN AT THE END : a closing-lane driver who reaches the first sign in queuing traffic (below about 37 mph) stays in its lane to the cones and merges there. One who reaches it in flowing traffic moves over from the sign, as with EARLY (rule 134 recommends merging in turn only at very low speeds), unless the traffic it is in starts queuing before it has found a gap, when it carries on to the cones.
EARLY : every driver in the closing lane moves over from the first sign, 800 yards out.
90% EARLY : the same, except that 1 driver in 10 carries on to the cones and merges there (the page's own figure).
LANE POLICE (0 to 20% of the cars, with any method): a driver who finds itself in the closing lane near the works keeps level with the queue beside it, so nobody behind can use the closing lane to get past, and in either lane it won't let anyone in ahead of it. Real ones sometimes straddle the line; here they sit in the lane.
With every method, nobody moves out into the closing lane from 50 m before the first sign onwards: drivers merging late, IN TURN or as lane police are drivers who were already in that lane. Past the first sign, once the traffic is queuing, nobody changes lane by choice; only drivers who have to leave the closing lane move over. That is the page's own simplification. The Highway Code's rule 288 for roadworks says "Do not switch lanes to overtake queuing traffic", but it doesn't forbid every lane change after a sign.
What the Highway Code says: rule 134, "Merging in turn is recommended but only if safe and appropriate when vehicles are travelling at a very low speed, e.g. when approaching road works or a road traffic incident. It is not recommended at high speed." Rule 288: "Where lanes are restricted due to road works, merge in turn."
Measured over the ten traffic patterns, from minute 5 to the end of the run: Queue , how far back from the end of the cones the last car below 25 mph is, averaged second by second, with the longest in any of the ten runs (marked ≥ if the queue ever reached the start of the road, when cars waiting to get on make it longer than shown); Through , cars an hour past the end of the cones; Lost , the time lost by the timed group getting to the end of the cones, against the time at its own wanted speed slowed to 50 from the signs; Lane use before the cones; and fairness. For the time lost, each run carries on after its end, with traffic still arriving but nothing else measured, until every timed driver is through (for up to 15 minutes more), so it is a measurement, not a minimum. Far beyond what the works can carry, when the timed drivers couldn't all get through in those 15 minutes, the run doesn't carry on: anyone still queuing counts with the time lost so far, and RUNS SO FAR marks the figure ≥, a minimum. For fairness each driver's place in the order passing the first sign is compared with its place passing the end of the cones: Jumped 10+ is the share of drivers who drove on in the closing lane past the sign and got ahead of 10 or more drivers who were in front of them there, and Most gained the biggest such gain in a run (averaged over the ten).
In this model, with two lanes down to one, 1,800 cars an hour (more than the works can take) and 30 minutes: IN TURN kept the queue shorter in all ten patterns, about a quarter shorter on average (0.6 miles against 0.8 with EARLY; the longest in any run 0.9 miles against 1.2). It got as many cars through (about 1,380 an hour either way), and its timed drivers lost about the same time, 289 s each against 283 s (more in some patterns, less in others). Nobody jumped the queue with EARLY; with 90% EARLY 5% of drivers got past 10 or more others, and the biggest gain in a run averaged about 27 places, for about the same time lost (284 s). IN TURN came out least fair here: about 1 in 6 of its drivers gained 10 or more places, because in this model the closing lane's queue moved faster than the open lane's between the sign and the cones (fewer drivers stay in it). Real drivers can choose the faster queue before the sign; the model's drivers don't keep the two queues level, so treat that figure with care.
The answer changes with the assumptions. With a 0.3 s longer gap in the works lane instead of 0.6, the cones can take more cars than the merge in turn feeds them in this model: IN TURN's queue was again shorter, by about a quarter, but it got about 100 fewer cars an hour through (about 1,470 against 1,570) and its drivers lost more time (176 s against 140 s, more in 9 of the 10 patterns), with 21% jumping 10+ places. At 1,500 an hour the queue barely reached the sign: there was little to choose (67 s against 66 s; the queue about 8% shorter IN TURN), and with the 0.3 s gap there was no queue at all. At 1,200 an hour the methods were the same (about 13 s lost) and nobody jumped. 10% lane police at 1,800 an hour cost time with every method (IN TURN 310 s, EARLY 304 s, 90% EARLY 304 s); queue-jumping IN TURN fell to 5%, and with 90% EARLY to 1%. The steady result is the queue: once it reaches back past the first sign, using both lanes to the cones makes it shorter, for about the same time through the works. The Minnesota Department of Transportation reports the same pattern from its studies: in slow, congested traffic the zip cut the length of the queue by up to 50% (40% was common) and made no difference to the time taken through the work zone, and in light, fast traffic drivers should merge early.
Not modelled: a separate parallel deceleration lane before the slip road (drivers here slow on the slip road itself), smart motorway lanes and variable speed limits, the queue at the roundabout at the end of a real exit slip road, drivers who signal and are let in early, vans and coaches, weather; at roadworks, closures of lane 1, works vehicles, lanes narrowed but not closed, drivers straddling the line, and the extra risk of any manoeuvre (the page counts time only). Distances on the road are exact; cars and lanes on the cameras are drawn wider than they really are so you can see them.
FURTHER WATCHING AND READING
CGP Grey: Traffic Has a Perfect Solution. Humans Are the Problem. , 5 min. Why human reactions turn small disturbances, like a car pulling in, into waves.
traffic-simulation.de: an interactive traffic simulator , 10 min to play. Martin Treiber, co-author of the IDM and MOBIL, lets you try on-ramps, lane closures and politeness yourself.
The Highway Code: Motorways (rules 253 to 274) , 10 min read. Joining, lane discipline, lorries in lane 3, overtaking and leaving.
Minnesota DOT: Late merge, the zipper system , 5 min read. A road authority's fact sheet on when late merging at a lane closure helps, and when to merge early.