Embedded Digital System                             

 

Smart toys revel in mobility, speech, control By Stephan Ohr, EE Times November 17, 2002 (2:18 p.m. EST) URL: http://www.eetimes.com/story/OEG20001201S0058 The future of smart toys may be glimpsed in a robotic animal that is billed as "autonomous, sensitive to its environment, able to learn and mature." The latest version of the Sony Aibo, the ERS-210, announced in October, represents an outrageous use of analog and mixed-signal technology. The little critter has a built-in stereo microphone and will learn and respond to about 50 voice commands. A stereo amplifier and speaker allow it to produce tonal sounds. Sensors on its head, chin and back simulate a sense of touch, while a camera in its nose replicates sight. Some 20 motorized joints on the dog realistically mimic walking, sitting, lying down, flapping its ears and wagging its tail. Its eyes are blinking LEDs, and their expression is described as "soulful." The robot's simulated emotions include anger and fear, surprise, dislike, sadness and joy. A $90 software package allows the robo-pup to "learn"-that is, to associate these emotional responses with certain objects, sounds or user responses. The $2,500 Sony Aibo is undoubtedly the ultimate plaything, a rich man's indulgence. But smart toys-including Furbies, Talking Barbies, radio-controlled blimps and racing cars, pocket games and video consoles-are growing up and morphing into a whole host of consumer electronic entertainment devices, boxes and consoles. Apart from smart processors (with ever more Mips and megaflops) and CMOS ASIC integration, which shrinks the size and cost of the electronics, essentially three additions can be made to enhance the value of smart toys. One is speech; another is remote control; and a third is movement. And while contributors to this week's Signals section may offer new options from which Santa may choose, they remain firmly practical when it comes to cost and functionality. Contributors who focus on speech will tell you that the attention span of any particular child for any particular toy is unfortunately limited. Even in these days of fading affluence, a parent doesn't want to spend $70 or $100 for a talking toy that is used only once. But the more realistic the speech capability, the more valuable the toy will be to its user-and the longer its longevity as a plaything.

Iguana Robotics' Anthony Lewis has a new angle on the kind of computations required to power a robot: using algorithms derived from human biology to control robotic devices. Embedding these algorithms on-chip offers more streamlined control than the traditional lookup table.

Speech in Toyland nonetheless represents a paradox, according to contributor Larry Gaddy, marketing manager for Winbond Electronics Corp. (Hsinchu, Taiwan). The big issue in a command-and-control applications environment-even a "you-talk/it-talks-back" environment-is not the processing power of the microprocessor or DSP used to decode speech, but rather the amount of memory. More memory means more realistic speech, but it also elevates the cost of the electronics. "For speech in toys, the cost issue is always brutal," says Gaddy.

Winbond, whose customers include automotive console makers like Blaupunkt in Germany, has developed industrial-strength speech recognizers-hands-free phone dialers for cars, and voice-activated speaker phones.

The key to conserving expensive memory usage, Gaddy believes, is to build recognizers that rely on acoustic models-not word models. Word-based recognition, under the best of circumstances, is only about 95 percent accurate. Acoustic models, particularly for digits, are 99 percent accurate, he says. The problems of recognizing speech in young children, whose pronunciations can be challenging as well as cute, may affect the accuracy of these smart toys.

"In toys, cost is king," according to contributors Jeff Rogers and Erik Soule, marketing analysts at Sensory Inc. (Sunnyvale, Calif.). "Typically, there is a four- to five-time mark-up in the toy industry, which means if technologies initially cost $5, this quickly translates to $20 to $25 retail for just the electronics. Once the plastic, plush, manufacturing and shipping are added, the product now retails at $50 to $100, or more."

Sensory conserves costs by using an 8-bit processor and customizing it with digital filters and special-purpose cells. Its products include speaker-dependent and -independent recognizers, speaker verification (voice password) systems, word-spotters, speech synthesizer, voice record and playback, and four-voice music generators. Target markets, they say, are Tomogochi products, voice-controlled robots, interactive pets, dolls, password-protected products and AIBO-type dogs.

The issue of adding radio-control modules to toys is addressed by distributor Insight Electronics in (San Diego, Calif.). Schlacht describes some of the transmitter-receiver modules readily available from companies like Infineon Technologies AG and RF Micro Devices. These enable serial data communications between controller and remote modules, but, admittedly, may be a bit pricey for use in toys. Until then, a common approach to a low-cost RF transmit/receive link is to use a completely discrete solution. On the transmit side, the simplest way to start is to use a known oscillator circuit and vary one or more quiescent parameters that will vary the transmitted power level. The amount of amplitude variation depends on the application, the environment, the required range and noise level tolerated. But 6 to 10 dBm is usually sufficient for controlling toys across a room or playground, Schlacht says.

Infrared (IR) is useful as a very low-cost solution, Schlacht acknowledges. IR's drawback is that it depends on an unimpeded line of sight (LOS) between the controller and the remote. "Any application involving an 8-year-old that requires him to maintain an LOS between his toy and his controller is doomed to frustration and an eventual loss of interest."

Iguana Robotics Inc. (Mahomet, Ill.) president Anthony Lewis and consulting scientist Ralph Etienne-Cummings, an assistant professor at Johns Hopkins University (Baltimore), have a particular slant on the kind of computations required to power a robot, whether it is used as an extension of a NASA space shuttle arm, or kid's toy. A moving device must effectively "see" where it's going, they argue, and this process of relating what they see to how they move is effectively a biological relationship. The devices they've created for commercial exploitation and for university research make a point of integrated "brain-style" object recognition.

The chips use what Lewis and Etienne-Cummings call embedded biomorphic technology (EBT)-algorithms derived from human biology but used to control robotic devices. Embedding these algorithms on-chip provides a more streamlined method of control than the lookup-table sequences used by microprocessors in their instruction-fetch/instruction-execute mode of operation. A recent chip built at Johns Hopkins, for example, can process the equivalent of 12 Gflops/milliWatt with roughly 5 bits of resolution.

Two chips serve as test pieces for EBT. One is an intelligent visual sensor that incorporates mixed-signal, "computation on read-out" (COR) that analyzes pixel data from a 128 x 64 array. The color processing circuits construct a histogram-effectively sizing these images-at a very high frame rate. It effectively performs an intelligent object recognition (albeit fuzzy) at a very high rate of speed. This was developed exclusively at Iguana Robotics.

A second chip, developed by Iguana with Johns Hopkins University, is a locomotion controller chip, capable of controlling the movement of a four-legged robotic toy animal. The current chip can generate the basic movement pattern for a small running biped robot. "While these techniques may seem exotic," write Lewis and Etienne-Cummings, "they may be the most effective route to building low-cost, sophisticated robotlike toys."

A number of the breakthroughs in motion control, image processing and speech recognition really depend on digital signal processing, points out senior fellow and resident visionary Gene Frantz of Texas Instruments Inc. (Dallas). The mass-storage industry depends on DSPs to control hard-disk drive read-head positioners, and thus increase disk drive density. DSPs also decode multimedia streams for digital versatile disk (DVD) players. But they also powered kids' toys like Speak-'N-Spell and the Julie doll.

The TI fellow details future improvements in both architecture and manufacturing processes: He predicts a $5, 5,000-Mips processor that consumes 0.1 mW/Mips by 2002; and a 50,000-Mips device that will cost just 15 cents and run on 1 microWatt/Mips by 2012, ensuring an increase in smart toys and entertainment appliances.

Meanwhile, the biggest job of the set-top box is MPEG-2 digital video decoding, points out Jeff Haight, product manager for high-end processors at MIPS Technologies Inc. (Mountain View, Calif.).

But typically the set-top box (STB) and the television reside in a home-entertainment environment that includes advanced audio processors, digital camera hookups and multimedia Internet downloads. It will only be a matter of time before a clever marketing executive insists on providing these services from the same box-if not on the same signal-processing device.

TOP  HOME