WASHINGTON -- A government office is sponsoring a robot race through the Mojave Desert with a $1 million cash prize, trying to build lasers powerful enough to down missiles and studying the feet of geckos. And it calls its best employees "freewheeling zealots."

DARPA -- the Pentagon's Defense Advanced Research Projects Agency -- is unlike any other federal agency.

New assignments in terrorism and high-powered computer software have brought unwanted attention and controversy to the office of just 240 employees. Before that, it had quietly transformed not only the U.S. military, with stealth aircraft, Predator drones and precision bombs, but also American society, with the Internet and global positioning satellites.

The man whose work generated much of the recent publicity, former Iran-Contra figure and retired Adm. John Poindexter, left the agency Friday after 20 months.

This week, House-Senate conferees will debate a Senate-passed provision to halt spending on his Terrorism Information Awareness project, which was designed to search the commercial transactions and personal records of millions of Americans to detect by computer the telltale patterns of activity by would-be terrorists.

Born in 1958 when the Soviet Union beat the United States into space with the Sputnik satellite, the agency is supposed to prevent technological surprises from U.S. adversaries. It long ago handed off ballistic missile work but still focuses on research that is over the horizon in time and space.

"We look at what future commanders might want," agency director Tony Tether says, "changing people's minds about what is technologically possible."

Too risky

DARPA investigates ideas "the traditional R&D community finds too outlandish or risky," its Web site boasts.

"If they aren't failing more than 50 percent of the time, they are probably too conservative," says Steven Aftergood of the Federation of American Scientists.

Washington University's Ron Indeck, an engineering professor who's worked with DARPA, says, "They go for very high-risk, high-payoff projects."

And the Poindexter effort was high-risk in every way -- as well as a classic illustration of how DARPA works.

With special flexibility to hire and pay competitive salaries, DARPA mines the defense contractors, universities and military offices that perform all its research.

"They drag in people as program managers who can have the greatest impact as fast as possible," Indeck said. "Nobody stays more than three to five years."