A Coin Flip Made This Route-Finding AI 90% Faster

Every time you order something online, a computer somewhere is figuring out the cheapest, fastest way to get it to you. That kind of planning is brutally hard — the number of possible routes explodes the moment you add more stops. So engineers use shortcuts: instead of hunting for the perfect answer, the software looks for one that's "good enough," usually within a set percentage of the best possible. That trade-off is what lets your delivery app respond in seconds instead of hours.

A team of researchers from Vietnam published a new paper describing a small but clever tweak to this process. Their software normally follows its own best guess, over and over, which sounds sensible but has a flaw: it can get stuck. Picture a driver who keeps circling the same neighborhood because that's where the map says the road is shortest. The researchers added a coin flip. Most of the time the program follows its best guess, but occasionally it deliberately checks the currently worst-looking option — which often reveals new, better paths.

On standard test problems — sliding puzzles, pancake sorting, and the classic traveling salesman problem (finding the shortest loop through many cities) — the coin-flip version finished with roughly 90 percent less computer work in the situations where the old method stalled. When the old method was already efficient, the gains shrank, which the authors honestly report. They also tested a version that keeps improving its answer over time, and it beat every rival they compared it to.

The practical payoff is speed and cost. Less computing means faster answers and lower electricity bills for the data centers running logistics, chip design, and robot navigation. Don't expect a dramatic overnight change — this is a research result, not a product launch, and it only helps when the software is already struggling. But it's the kind of quiet improvement that eventually shows up in your delivery window, your flight connections, and your mapping app.