The ocean is an awesome place full of sharks, shipwrecks and strange undersea creatures. But it’s not all like that. How’s a robot to tell what’s unique
At Woods Hole Oceanographic Institution in Massachusetts, computer scientist Yogesh Girdhar is figuring out how to imbue bots with a sense of curiosity. He wants them to be able to filter out the more common features, and focus on the remarkable.
The curiosity software starts running the minute the robot is dropped into a new place. It has little information about what the world looks like initially, but slowly begins making sense of what it sees by searching for patterns in the data. In the ocean, those patterns represent things like sand, kelp or fish.
As the robot travels, it continues learning to recognise new objects, while watching out for things that don’t quite belong. That mysterious stuff is automatically tagged as the most interesting, piquing the robot’s curiosity and causing it to head over to take a closer look.
Testing the waters
Girdhar and his colleague, Gregory Dudek of McGill University in Montreal, Canada, tested the system on aerial and underwater photographs in the lab, but weren’t sure how it would fare in the real world. So they programmed it into AQUA – a flippered robot designed to explore shallow-depth attractions such as coral reefs and the undersides of ship hulls. They brought it to Barbados and set it loose in the open ocean.
Underwater, AQUA looked “like a puppy”, Girdhar says, racing and eagerly sniffing at new or unusual sights. In one set of tests, it successfully wandered over to check out nearby sea plants and corals, spending less time on bare patches of sand.
In another test, a scuba diver swam by the robot to see if they could catch its attention. It noticed and followed the diver for some time, hovering nearby when they stopped.
The researchers plan to try the algorithm in more places and situations, and with other types of ocean-roving robots. They think it will be useful for when maintaining reliable communication with an exploring robot is hard, such as in remote parts of the sea or on another planet. That way, the robot can make some decisions for itself about what to check out, and send the most interesting information back to its owners.