Monday, August 27, 2018

What Could Possibly Go Wrong?

Students at Worcester Polytechnic Institute design autonomous, spear-toting robot to drive invasive lionfish away from coral reefs.

3 comments:

G. Verloren said...

Uhhh....

A fine theoretical concept, but there is no earthly way that image recognition software currently exists which is sophisticated enough to reliably and successfully identity lionfish and only lionfish.

John said...

He said, as the robot pursued him relentlessly through the murky waters, spear tips glinting creepily in the occasional flashes of sunlight. . . .

G. Verloren said...

That's kind of my point.

The state of image recognition in the world today is such that I am confident the proposed robot would have a very poor performance record due to constnatly acting on false positives - mistaking one species of fish for another, mistaking a pattern in the sand of the sea floor for a fish, mistaking the patterning on a diver's wetsuit for a fish, et cetera.

There are so many factors that could contribute to insane numbers of false positives. Water clarity due to sediment, current light levels, turbulence and currents, fish being imaged from unsual angles or being obscured by other fish, mud or debris on the camera lens, hardware or software failures causing the units to go on the fritz and spear things at random...

Speaking of hardware failures, how do you keep the moving parts maintained? If you're spearing fish, how do you remove the fish which will inevitably get stuck on the spear? If you miss and imbed in a rock, how do you get the spear loose? What happens if the robot gets tangled in a fishing net, or randomly hooked by a fishing pole?

What happens if a shark comes along and is intrigued by the robot, does the usual form of shark investigation where they bite a foreign object to probe it, and the spear actuates in the process and impales the shark, the shark attempts to flee, and now maybe the robot gets dragged several miles into a different environment that it's not supposed to be in, and starts killing local fish it shouldn't be killing once it gets loose, because it was never programmed to exclude that species since it was never meant to be in proximity? Or heck - maybe the shark just dies, falls on the robot, and pins it to the sea floor.

There are so many points of failure here that make this idea of autonomous spear robots seem utterly insane.