MIT's Robotic Fish Takes First Swim


CAMBRIDGE, Mass.--Consider the fish: highly maneuverable and an
effortless swimmer, this animal 160 million years in the making is
superbly adapted to its watery environs. Now, in work that could lead to
mini submarines with similar attributes, MIT engineers have developed
the first robotic version of Nature's piscine wonder.

In mid July the researchers' creation, patterned after a bluefin
tuna, took its maiden swim down the MIT Testing Tank. That swim and
others since have been flawless, reinforcing the engineers' belief that
the Lycra-sheathed robot could become an important tool toward
understanding the physics of swimming and more.

The "robotuna" project began about three years ago with the overall
goal of developing a better propulsion system for autonomous under-water
vehicles, or AUVs, said Michael S. Triantafyllou, a professor of ocean
engineering who is leading the research team.
AUVs are small robotic submarines that have great potential for
mapping the ocean floor, finding the sources of underwater pollution,
and more. Yet currently they can stay under-water for only limited
amounts of time, restricting their usefulness. "You simply can't put
enough batteries on board for long-term missions like exploring the mid-
Atlantic ridge for a couple of months," said David S. Barrett, a
graduate student in ocean engineering who is developing the robotic fish
for his doctoral thesis.

The solution lies in better batteries or a better propulsion
system, Mr. Barrett said. "And since no one wants to put a nuclear
device on an autonomous vehicle, we're going the propulsion route." Fish
have the most efficient propulsion system around, so the researchers
decided to copy Mother Nature.

The result of their efforts, which included hours of watching tuna
at the New England Aquarium and combing the literature on how fish swim,
is Charlie the Testing Tank Tuna. About four feet long, Charlie was
designed to resemble the real thing as closely as possible. Hence the
robot's 2,843 parts include over 40 ribs, a set of tendons, a segmented
backbone with vertebrae and, of course, its Lycra skin. The robot swims
down the tank propelled by a tail that gently swishes back and forth as
its flexible body follows suit.

Charlie is the first of a series of "robofish" that the engineers
plan to build, each with successively greater abilities. For example,
the current fish is attached to a structure that guides it down the tank
and contains all of its electronics. But the "robopike" that the
researchers plan to build next will probably be connected only to a
tether, allowing the creature to make hairpin turns.

The ultimate goal? A fully autonomous fish that "we could throw
into Boston Harbor, tell to go somewhere, and have it come back," said
Mr. Barrett. The engineers are hopeful that such a fish could be a
reality in about five years.

For now, the researchers are concentrating on getting the robot to
swim as efficiently as possible in a straight line. "We hope to find out
if [our system] is or is not better than conventional propulsors," Mr.
Barrett said.

Although a new, highly efficient propulsion system for AUVs is the
main goal of the robotuna project, Mr. Barrett stressed that the robot
is a "multi-mission experimental unit" with a number of other scientific
thrusts.

For example, the researchers are learning more about the
fundamental fluid mechanics of swimming. Further, the robot is a test
bed "to let us check out a variety of controllers and sensors for
systems like this."

In a specific example, the researchers are planning to "evolve" a
computer control system that will make the robot swim most efficiently.
Using the computer, they'll create models of different control systems.
Each system will be represented by a chromosome-like data structure
whose "genes" dictate how the robot moves.

The researchers plan to test each control system through Charlie
using a genetic algorithm to determine which will make the robot swim
most efficiently. Just like in natural selection, then, "the top 60
percent will get to have virtual offspring," said Mr. Barrett,
recombining the "genes" to produce even more efficient control systems
(the bottom 40 percent effectively goes extinct).

"So the overall efficiency gets exponentially better and better,
and we end up with an arrangement of control parameters that make for
the most efficient run," Mr. Barrett said.

Key to the project are several students. They are: Pehr Anderson, a
junior in electrical engineering and computer science (EECS) who is
developing sensors for the robot; John Kumph, a senior in mechanical
engineering who oversees robot control; Gillian Lee, a senior in EECS
who is developing the genetic algorithm; Elise Martin, a junior in
mechanical engineering who is developing the skin of the robot; Scott
Miller, a graduate student in mechanical engineering who is working on
the hydrodynamics of the tail fin, and Owen Wessling, a senior in EECS
who is developing the data acquisition system.

The work is funded by the Advanced Research Projects Agency, the
Office of Naval Research, the MIT Sea Grant College Program, the Woods
Hole Oceanographic Institution, and MIT's Undergraduate Research
Opportunities Program.


Topics: Artificial intelligence

Comments

Back to the top