Behind the televised clatter of mechanized combat during last month's RoboGames in San Francisco was a quiet concern. Robots won't save the world. But, equipped with enough intelligence and mechanical precision, they could disarm bombs and perform battlefield surgery. They could weld car bodies, explore Mars or Titan. The concern, expressed by proselytizers like David Calkins of San Francisco State University or Bob Bluhm of Tektronix Inc., was over who would take responsibility for developing this machinery.

In the absence of a large-scale multicorporation effort, robot development seems to fall on the shoulders of students and hobbyists. Competitions like the RoboGames and the more-visible Darpa Grand Challenge seem designed to encourage student participation. The Defense Advanced Research Projects Agency, in fact, is offering a $1 million prize to the university team that successfully sends an autonomous vehicle across the Mohave Desert, roughly 140 miles, from Barstow, Calif., to Las Vegas. Tektronix is sponsoring the Cornell University team, said Bluhm, who is general manager of the company's "value scope" division.

In San Francisco, local television made much of the clattering, spark-filled robot combat duels between spinners and wheeled platforms. Of less interest to the cameras were contests like Robo-Magellan, in which an autonomous overland vehicle must find its way through an unknown, obstacle-filled territory-frequently pausing to recalculate and sometimes losing its way and stalling. Robo-Magellan, like the Darpa Grand Challenge, failed to produce a winner. "The technology development stage is probably between that of an Apple I and an Apple II," said Calkins, who is president of the American Robotics Society and professor of robotics and computer engineering at San Francisco State, where the RoboGames were held.

"Sony and Honda have built brilliant walkers," Calkins said, referring to the cost-is-no-object biped androids whose mechanical movements emulate human motion. "But these guys," he said, referring to contestants in the games, "these guys build them in their garages for one or two thousand dollars."

One reason is that market applications may be too fragmented to encourage multicompany participation. Unlike personal computers or cell phones-which require not only hundreds of millions of units, but also a massive infrastructure-the market for robots is splintered among many specialized applications (like autonomous vacuum cleaners or lawn movers), each requiring its own specialized programming effort, if not its own specialized hardware. No single market sliver seems large enough to encourage billion-dollar investments. Only 42,000 industrial robots are consumed each year, according to some estimates. Even with a digital-to-analog converter in every joint of a six-axis robot, there wouldn't be enough volume to pay for the laser trimming you would need to build a 14-bit-accurate device.

Nevertheless, it is possible that students, hobbyists and other early adopters will be the ones to show the way.

Three skill sets go into the construction of a robot, Calkins said. One is electronics intelligence, involving the inter-pretation of visual sensor data. Another is electromechanical, largely motor drivers and servos. A third-most dramatic in the combat machines-is precision welding and machining.

On the electronics side, the sensory inputs on the autonomous robots competing in the RoboGames were either CMOS image sensors or infrared detectors. A crude visual pattern-recognition process would determine what the robot was looking at and how it should respond. In one RoboGames event, Aibo Soccer, Sony's robot dogs were programmed to distinguish an orange ball on a green playing field, and to swat it toward a color-coded goal with their snouts. The smarts for visual pattern recognition were provided by a MIPS R6000 processor. Microsoft Corp. provided C++ programming tools in sponsorship of one of the student teams.

Stephan Ohr (sohr@cmp.com) is technology editor for the EE Times Network and site editor for planetanalog.com.