In simulations, hybrid model makes unmanned searches 15 times as fast

A hiker is lost in a forest, winter weather is closing in, and an autonomous robotic drone is sent to search. The route cannot stay fixed for long: each image or new piece of information may change where the vehicle should go. Research led by Rohan Ghuge of the University of Texas at Austin now puts a number on a middle-ground strategy: in computerized simulations, two adaptive rounds were 15 times as fast as a fully adaptive model.

The choice is between speed and responsiveness. A fully adaptive vehicle recalculates its path after every new piece of information, which can produce better routes but consumes more time and resources. A nonadaptive vehicle sets a route once and does not divert, even when new data suggests it should. Ghuge worked with Rayen Tan and Viswanath Nagarajan of the University of Michigan to test a hybrid between the two.

The hybrid vehicle follows a designated course and waits until the end of each course—defined by the researchers as a “round”—to recompute its path. The first round is nonadaptive; later rounds become increasingly adaptive. In the simulations, two rounds cost only 12% more than a comparable fully adaptive search. After the first three rounds, the researchers found that accuracy did not increase markedly.

So what changes in practice? A search-and-rescue operation could trade a modest cost increase for a faster response, while a utility company could use the same balance when locating the source of an outage. Fewer replanning steps could also ease logistical pressures such as battery life and how often the vehicle uploads data. The researchers’ result is not yet a field demonstration: it comes from computer simulations.