Bats do it, dolphins do it, and now even a robot in Connecticut can do it: use sonar to identify objects accurately. And while the robot can't fly or swim, echolocation and the ability to learn like those mammals help it recognize objects as small as 1 millimeter and distinguish between the heads and tails of a coin.

"Sonar is the most common sensor used in robotics," said Roman Kuc, the director of the Intelligent Sensors Laboratory at Yale University in New Haven, Conn., "but it's a low-level sensor" usually used just for obstacle detection. "Obviously bats and dolphins do a lot more with sonar, so I tried to determine how they work with it."

Eventually Kuc developed the Rodolph, or robotic dolphin, a machine built around several design principles found in nature. The robot's head, for example, is configured like that of a bat, but in the machine's case, the central mouth (the transmitter) and two ears (the receivers) are actually 60-kilohertz Polaroid transducers. In addition, the Rodolph's ears can rotate toward the echo's source; most other sonars are fixed in their configuration.

The robot also moves the sonar so an object is positioned with the sound centered between the ears, Kuc said, providing "a unique set of echoes from which an object can be identified." The transmitting/receiving device is placed at the end of a robot arm, which moves approximately 10 centimeters above the work surface. Looking down at a 45-degree angle on the surface, the device transmits a ping or click. At that angle, "most of the sound goes away from the surface," he said, "and you only get information-rich echoes coming back."

Kuc also followed an ear cochlear model to replicate how the mammalian ear extracts important information from the echo waveform.

Using these principles and an ultrasound wavelength of 1/2 centimeter, the robot can tell the difference between ball bearings 1 millimeter in diameter as well as between a coin's heads and tails. "The ability to differentiate objects is completely dependent on signal-to-noise ratio," Kuc said. "If you have clean signal, you can tell a lot."

Such improvements are possible because the robot can move its head around an object like a pet investigating a strange new toy. That motion places the object along the transmitter axis, "so you know the kind of echoes you're looking for," Kuc said, "increasing the intensity of the incident acoustic pulses [to] get a big echo." In addition, when the ears rotate, the bandwidth of echoes is maximized, enabling the robot to identify and differentiate objects.

To help perform those tasks, the robot uses another animal ability: learning. Through a memory model that stores only novel echo features, the Rodolph can detect even slight differences between objects. "It's just like a mother dolphin telling her baby that something up ahead is an edible fish," Kuc said. "Mother pings the object with echoes, the echoes come back, and the baby understands" what it is. The robot goes through a similar learning phase to build up a database, which it uses to compare sometimes almost identical echoes.

The memory model proved especially valuable during the Rodolph's development. According to Kuc, as the robot arm extended, the elevation of the head varied up to 2 degrees, decreasing accuracy. Researchers decided to store different echo features at different elevations, scanning objects in the learning mode in 0.25-degree increments from 42 to 48 degrees. "Once you get those echoes, almost any small deviation from the [optimal] 45-degree angle gets you the signals you want to see."

The sonar system, Kuc added, holds a number of advantages over a camera-based sensor system. For example, the ultrasound transducers produce two one-dimensional signals, which are far faster to process than a camera's fully 2-D signal. Kuc's system saves time because it doesn't produce an image of the object or environment for a human viewer; instead, he and the Rodolph look at an echo's waveform.

Kuc is working on several applications for this system, such as in an underwater vehicle. He is also working with Masashi Kawasumi of Tokyo Denki University on a system for the disabled—such as for quadriplegics who now use puff-and-sip chairs or voice-recognition systems—to improve communications through lip reading. Such a device would focus on the subject's nose as a repeatable location, then register lip motions that generate a variety of echoes. "We're looking at how many facial gestures we can differentiate, what is the data rate a face produces, and the like," Kuc said.

Another project involves research into how animals perceive large objects, so a sonar-based robot can do the same. "We're building a mobile robot arm that can explore the environment, forage for food, and recognize the boundaries of its environment," Kuc said. Mother Nature shouldn't be afraid of the competition, however: The Rodolph will probably never be able to jump through hoops at an aquarium show.

home | features | weekly news | marketplace | departments | about ME | back issues | ASME | site search

© 1998 by The American Society of Mechanical Engineers