Robot cars versus phantom traffic jams

Thursday, May 18th, 2017

A new study suggests that even a few autonomous cars could ease congestion for everyone:

You’ve likely seen the demonstration of phantom traffic jams where cars drive around in a circle to simulate the impact of a single slowing car on a road full of traffic. One car pumps its brakes for no particular reason, and the slowdown ripples through the traffic. Now, the University of Illinois research, led by Daniel Work, shows that placing even just a single autonomous car into one of those circular traffic simulations can dampen the effects of the phantom traffic jam.

The team’s results show that by having an autonomous vehicle control its speed intelligently when a phantom jam starts to propagate, it’s possible to reduce the amount of braking performed further back down the line. The numbers are impressive: the presence of just one autonomous car reduces the standard deviation in speed of all the cars in the jam by around 50 percent, and the number of sharp hits to the brakes is cut from around nine per vehicle for every kilometer traveled to at most 2.5 — and sometimes practically zero.

Because fuel use increases when when cars slow down and have to get back up to speed, the presence of the autonomous vehicle also reduces fuel consumption. According to the calculations by the team, in fact, the savings is as much as 40 percent when averaged across all the cars in the traffic flow.

It’s interesting that these improvements can occur even with a single vehicle in a flow of 20 other cars. And it’s also worth noting that the level of autonomy required to have this effect isn’t the kind that Waymo, Uber, and others are seeking to build — it’s more akin to the adaptive cruise control already featured in many higher-end cars.


  1. Bob Sykes says:

    The Antiplanner has a number of posts on autonomous vehicles, and he thinks they will make deep inroads into public transit, and in some cities a significant part of the public transit system will disappear. The key is that autonomous vehicles greatly increase the capacity of highway lanes, and the costs would be substantially smaller than our highly subsidized (>80%) light and heavy rail systems.

  2. Candide III says:

    Autonomous vehicles greatly increase the capacity of highway lanes how? Large increases in occupancy aren’t going to happen, I’ll warrant, and minimum inter-car distance is going to be determined by safe braking distance no matter how clever the car is. (Imagine it were set to say half that and a box falls out of a truck on a busy highway? Ugh. A single pile-up would be enough to sue the maker into oblivion, and rightly in my opinion. Shit happens.) Also, road infrastructure in low-density areas is heavily subsidized too, so it’s a bit of a choose-your-poison type situation.

  3. Eli says:

    Change doesn’t have to be framed in the context of how its always been done.

    The Feds tax gasoline, and its in the range of twenty-five cents a gallon for gas at the pump. This goes to the DoT for highway infrastructure. This is used to “match” funds put up by State DoT’s (State gas tax of fifty cents and more) to maintain Interstate highways. But lets not call it a subsidy, right?
    Getting tractor-trailers off the road would be a key to decongesting commuter routes as well.

    Every city has package shipping hubs from freight movers to FedEx, UPS, etc. These operators have units running the same route in the same direction at the same time every day from/to the same depots every day.

    If the rail system were set up like the Eisenhower Interstate system (not owned by a company/monopolized by a company, rails were built on Federal rights-of-way) and supported with similar tax scheme, fuel consumption and road congestion would be reduced all around. This would reduce the rate at which Interstates are being expanded as well.

    And for the same reason autonomous cars can reduce traffic jams — uninterrupted flow in one direction.

    Rail operators would buy access to segments of rail in much the same way airlines buy access to runways and terminals.

  4. Candide, my understanding is that safe braking distance is determined largely by human reaction times. Presumably an automated system could react much, much faster and thus the safe braking distance could be reduced to near the physical limits of the car’s brakes.

    Even if you left substantial “buffer” distance on top of that I could still see it substantially reducing inter-car distance on the highways, especially given that non-optimal human reaction times (i.e. your average somewhat-distracted commuter, not a fighter pilot) are rather long.

  5. Bruce says:

    The Ford executive who puts a microwave radar that brakes or just beeps for deer in F-150s will make money.

  6. Candide III says:

    Scipio, human reaction time is a minor factor. Under typical “bad” conditions (slippery road, almost-negligent human driver) coefficient of friction is taken to be 0.25 and perception-reaction time 2.5 seconds ( Assuming car velocity on highway to be 100km/h≈30m/s, the braking time will be (30m/s)/(0.25*9.8m/s²)≈12s and the braking distance will be (30m/s)²/(2*0.25*9.8m/s²)≈180m. The distance car travels until the driver floors the brake pedal is (30m/s)*(2.5s)=75m. If you take typical “good” conditions, the proportion is about the same. Even if you cut the driver-related component down to nothing you only win 30%, hardly a “great increase”.

  7. Isegoria says:

    On a highway filled to capacity (which is about 2,200 vehicles per hour per lane), about five percent of the available road space is taken up by cars:

    When traveling at 100 kph, if all vehicles on the road are simply equipped with adaptive cruise control, highway capacity can safely increase by a factor of 1.4. And if all vehicles on the road are equipped with both adaptive cruise sensors and communication, capacity can be increased by a factor of 3.7. And this increase is without any infrastructure modification: it’s purely from making our cars smarter with technology that is commercially available today.

  8. Candide III says:

    That’s the point: you can’t take up much more than those 5% and keep safe braking distances. Say the average car is 5m long, that means that at capacity the distance between cars is about 100m. If you increase density 3.7 times, the distance between cars will be a bit more than 3.7 times less, approximately 25m. That’s far below safe braking distance at 100km/h in good weather even if you ignore human perception-reaction delay. A highway full of vehicles at such density is a huge pile-up waiting to happen.

    It’s common in gadgetbahn to propose extremely tight headways, presuming computerized control allowing vehicles to behave as if they’re connected by a rod. Personal rapid transit proponents argue the same. In reality, such systems have been a subject of research for train control for quite a while now, with no positive results so far. Safety today still means safe stopping distances.[1]

    Meanwhile, a single direction of Tokaido Shinkansen routinely runs at 15,000 passengers per hour.

Leave a Reply