Snow plow route optimization · real road networks
Every street in Leadville, plowed.
A plow can't skip a street, so the only thing left to optimize is the driving in between. PlowPath solves that problem — capacitated arc routing — on real OpenStreetMap networks. The map is one of its solutions, replayed.
Every street has to be driven. The waste is everything else.
Snow plowing is an arc routing problem: a plow has to drive every street in its zone, so the only thing left to optimize is the driving in between. Driving with the blade up to reach the next unplowed street is deadhead. The deadhead ratio — blade-up distance over total distance — is the number PlowPath drives down, subject to priority tiers, salt capacity, depot returns and several trucks sharing one network: the capacitated arc routing problem.
Same streets. Same 2 trucks. 74% less blade-up driving.
Canton, Baltimore, Maryland, plowed two ways. A greedy fleet that heads for the nearest unplowed street drove 39 km blade-up (a 40.6% deadhead ratio); PlowPath's CARP solver, planning the routes first, drove 10 km (15.0%).
More trucks, more waste — unless they coordinate.
Swept from one truck to six in four towns, greedy's deadhead ratio climbs 14–25 points while CARP's moves -17–6.
28 cities. 16 states. 77,533 road segments.
- Leadville, CO — 529 road segments, best deadhead ratio 0.5%
- Oberlin, OH — 445 road segments, best deadhead ratio 0.7%
- Galena, IL — 843 road segments, best deadhead ratio 2.0%
- River Edge, NJ — 836 road segments, best deadhead ratio 3.5%
- Woodstock, VT — 844 road segments, best deadhead ratio 4.6%
- Saratoga Springs, NY — 3,280 road segments, best deadhead ratio 6.0%
- Ithaca, NY — 1,828 road segments, best deadhead ratio 6.7%
- Houghton-Hancock, MI — 1,620 road segments, best deadhead ratio 6.7%
- Northampton, MA — 2,376 road segments, best deadhead ratio 7.1%
- Marquette, MI — 2,037 road segments, best deadhead ratio 7.3%
- Elkridge, MD — 1,399 road segments, best deadhead ratio 8.9%
- Steamboat Springs, CO — 2,079 road segments, best deadhead ratio 9.0%
- Dubois, PA — 2,678 road segments, best deadhead ratio 9.2%
- Bel Air, MD — 1,088 road segments, best deadhead ratio 9.7%
- Erie, PA — 7,915 road segments, best deadhead ratio 10.3%
- Montpelier, VT — 991 road segments, best deadhead ratio 11.4%
- Brattleboro, VT — 1,398 road segments, best deadhead ratio 12.1%
- Canton, MD — 606 road segments, best deadhead ratio 12.5%
- Easton, PA — 1,703 road segments, best deadhead ratio 13.1%
- Burlington, VT — 2,215 road segments, best deadhead ratio 13.4%
- Portland, ME — 6,218 road segments, best deadhead ratio 13.6%
- Breckenridge, CO — 889 road segments, best deadhead ratio 14.4%
- Minot, ND — 5,418 road segments, best deadhead ratio 15.2%
- Keene, NH — 2,677 road segments, best deadhead ratio 16.3%
- Ellicott City, MD — 5,357 road segments, best deadhead ratio 18.0%
- Flagstaff, AZ — 6,896 road segments, best deadhead ratio 18.1%
- Duluth, MN — 10,200 road segments, best deadhead ratio 22.1%
- Juneau, AK — 3,168 road segments, best deadhead ratio 30.2%
One engine, sixteen stages.
OpenStreetMap networks classified into priority tiers; greedy, Chinese Postman, CARP and NSGA-II solvers; salt capacity, depot reloads, grade-adjusted fuel, storm accumulation and sensitivity sweeps. Walk the full pipeline.
Frequently asked
What is the deadhead ratio?
Deadhead distance divided by total distance driven — the share of driving done with the blade up. Lower is better. Below 20% is excellent; a fleet with no plan often wastes 40–60%.
How is this different from delivery routing?
Delivery routing is a node problem: choose which stops to visit and in what order. Plowing is an arc problem: every street must be driven. Salt capacity, depot returns, priority tiers and several trucks sharing a network make it the capacitated arc routing problem, which is NP-hard.
Is the data real?
Yes. Every network is pulled from OpenStreetMap and every figure on this page is read from the same exported runs the gallery shows. Where a run could not reach a street — usually a one-way segment cut off from the depot — the gallery reports the coverage it actually achieved rather than rounding up.
Can it handle snow that keeps falling?
Yes. A storm simulation models accumulation over time and re-plans when plowed streets fill back in past a threshold, tracking re-plow events and clearance time across the storm.
Is it only for snow?
No. The solver covers every edge of a network with capacity-limited vehicles that return to a depot. Street sweeping, waste collection, road inspection and meter reading have the same shape.
Plowing a town this winter?
Questions about the method, a town you'd like to see routed, or a fleet to plan for this season: send a note. Say how many trucks you run and where they load salt, and the reply can be specific.