For a few months, he wondered how he could say no to the Rockefellers.

It wasn't every day, after all, that they asked an immigrant from India to revive one of their companies, and Romesh Wadhwani was proud of that. The Rockefeller family's venture capital partnership, Venrock Associates, had watched American Robot Corp., its failing new company, run through a million dollars. After almost two years of operation, says Wadhwani, the company "had no product, virtually no people, and no marketing strategy."

Those lost years put American Robot far behind in a market in which too many competitors were pummeling one another with vicious price cuts.Wadhwani had run his own computer company before, but he knew nothing at all about robots. Nor did he know how to attack the big Japanese robot makers who, like champion sumo wrestlers, threw every U.S. company that took the mat against them. For a while, it seemed that the only way U.S. companies could deal with the invasion of Japanese robots was to sink cargo ships in the Pacific.

Undaunted, Wadhwani accepted the job. Today, three years later, he is not only still on his feet against the sumo wrestlers, but is also scoring some points of his own. American Robot's sales should reach $10 million in 1984, putting it among the top dozen or so U.S. producers, and Wadhwani projected a profitable fourth quarter, the company's first. He has even licensed the company's robot, called Merlin, to Daikin Industries of Osaka, which has agreed to sell at least 350 of them in the toughest market of all, Japan.

The wrestling match is far from over, though. Along with American Robot, more than 50 American and dozens of Japanese companies are struggling for a U.S. market that is still quite small. Although one analyst predicted a growth rate of more than 50% in 1984, the robot market in 1983 amounted to an estimated $170 million to $230 million, less even than Apple Computer Inc.'s quarterly sales.

All that competition for a small market has devastated the industry. For U.S. producers, dollars are walking out the door faster than robots: They had losses equal to 42% of net sales in 1982 and even more in 1983, according to the U.S. International Trade Commission. Only one U.S. company was profitable in 1983, and among the losers were the robot divisions of such giants as Westinghouse, Cincinnati Milacron, and GMF, a joint venture of General Motors and Fanuc, the Japanese robot maker. "I wouldn't be surprised if the number of U.S. robot companies was cut in half," says Gerald Michael, an analyst with Arthur D. Little Inc., the consulting firm. "Out of the 25 remaining, only 7 or 8 will have the lion's share."

Will American Robot be one of the survivors? Wadhwani won't succeed on the strength of his sales to Japan, or even on the capabilities, of his Merlin. He and other American robot makers can't compete with the Japanese in making the robot arm itself. Instead, they are changing the rules of the game to reward brains and agility rather than brute strength.

This strategy requires complex new software that makes robots smarter, able to "talk" electronically with other computerized equipment about what they are doing. It also requires teaming U.S. or Japanese robots with a variety of peripheral equipment, then selling the package as a complete system to perform an entire manufacturing task. "The day of the naked robot is gone," says William Rooney, vice-president of marketing at American Robot. Adds Laura Conigliaro, an analyst at Prudential/Bache Securities Inc.: "It's been a long time since I heard anybody say, 'We're just a robot company."

Ultimately, what is at stake here is nothing less than a share in building the much-heralded "factory of the future," an eerie place in which computerized controllers operate a network of robots, conveyor belts, and artificial-vision systems to produce goods with scarcely any human intervention -- machines capable of going through their paces even at night, with the lights off. "Robot," in fact, is something of a misnomer for what American Robot or any other company that hopes to compete in the future will have to produce. The robot of today is simply an automated machine tool; the robot of tomorrow promises to be a component of what may be the next major chapter in the Industrial Revolution.

If this ambitious software-based strategy succeeds, the downtrodden U.S. industry will capture the most profitable end of the robot business and will leave the Japanese in the unhappy role of commodity suppliers, providing the arms that make up only about 30% of the cost of an installed robotic system. The Japanese are responding by building more intelligent robots of their own, but they operate at a disadvantage under these new rules: Software has long been their weak point, and the installation of complete systems would require them to provide costly support services in the United States.

The man who is trying to make this strategy work at American Robot is a 37-year-old engineer who thrives on the frenetic tempo of high technology in this country. "I couldn't ever run a successful business in India," says Wadhwani. "It's not fast-paced enough." Decisive as well as energetic, he moves with the instincts of a hawk that has spotted quarry on the ground.

In 1983, to expand the capabilities of Merlin, Wadhwani checked out dozens of companies that were developing artificialvision systems. As the "eyes" of a robot, artificial vision utilizes sophisticated software to turn pictures into digital information, then uses additional software to recognize and analyze images and patterns. Wadhwani found the system he wanted at the University of Manchester, in England.

"We talked to the bursar and said we weren't leaving until we had a deal," recalls Wadhwani. "He was shocked -- he wasn't used to doing things that way." Seventy-two hours later, American Robot had 25% ownership in a university spinoff, Visual Machines Ltd. "He knew what he wanted," says Paul Gregory, the company's managing director. "There was no messing about with him. He's a very direct character."

Decisiveness was exactly what American Robot needed when Wadhwani took over. The company, whose brief life has mirrored the domestic industry's change of emphasis from hardware to software, traces its roots back to 1977 in Winston-Salem, N.C., where John Gallaher, then a 25-year-old computer programmer, started an enterprise in the back room of his father's real estate office. Gallaher Research Inc. put together computer systems for factories and offices.

A local businessman came to Gallaher in 1978 with a problem: The Occupational Safety and Health Administration (OSHA) was requiring him to spend $1 million to soundproof his yarn factory. Could Gallaher devise a less expensive way to do it? Gallaher built a remote-controlled manipulator arm that allowed workers to use the robot while stationed in a soundproof booth. "We had the prototype working well," he recalls wryly, "and then OSHA relaxed its rules."

To salvage the project, Gallaher turned the manipulator arm into a $1,000 home robot kit and sold two dozen of them. Later that year, a company asked him to modify the robot for use in a factory. With money put up by local businesspeople and a partner in a Wall Street investment firm, Gallaher had a new version of the robot ready for a trade show in 1980. Made out of plastic, it cost only $10,000.

Not long after the show, Gallaher's Wall Street patron helped him strike up a deal with Venrock. Although he was the brains behind the operation and had, in 1977, cashed in his $2,000 life insurance policy to invest in it, Gallaher says that he emerged from the paperwork with only 5% equity in the company, renamed American Robot. "I didn't know that as founder you didn't have to have only as much stock as you could afford," he says mournfully.

Five-percent control, Gallaher discovered, meant no control at all. The professional managers hired by Venrock to run the company insisted on dropping the plastic robot and switching to metal instead. That raised the robot's price and changed its applications from light tasks, such as loading bottles into cartons, to much more strenuous jobs.

Strongly disagreeing with this strategy, Gallaher left the company early in 1981. Soon after, the Venrock managers left, too. In a venture capitalist's nightmare, the start-up had disintegrated in less than a year. "There wasn't much of a company left," says Anthony Sun, a Venrock general partner. "We gave direction to the four remaining employees, all engineers, to finish development of a prototype."

In the fall of 1981, Venrock heard that Wadhwani was available. Born near New Delhi in 1947, Wadhwani was always a tinkerer who seemed destined either for jail or for the kind of high-tech job that Venrock was trying to fill. For a while, it looked like a close call.

At the Indian Institute of Technology, in Bombay, he and a few classmates built a rocket and got permission from the chief of police to fire it off in a stadium. "We had to dig a hole eight feet deep," he says, then turn the rocket upside down so that it would blast off into the ground. "They made us anchor the rocket, too. I kept saying, 'Where's it going to go if it's pointed into a hole in the ground?" He adds: "I'm not sure they understood Newton's law of action and reaction back then."

Another time, when he designed an intercom system, Wadhwani and a group of friends dug up a street to lay wire between two dormitories. "We didn't ask for permission," he says. "If we'd asked, somebody would've said no." He also says he built a soda vending machine that used a peculiar sort of gravity feed: When the machine dispensed one glass bottle, hundreds of others fell to the next slot in an uproar of clinking glass. "Coke hadn't seen anything like it," he chuckles. "And they haven't since."

After three years at Carnegie-Mellon University, in Pittsburgh, Wadhwani emerged in 1972 with a PhD in electrical engineering. "I had no clear idea of being an entrepreneur when I came here," he says. But he and an Indonesian classmate started their own company, called Compuguard, to sell computerized controls for heating, air conditioning, and elevators in office buildings. "Compuguard was my learning experience," says Wadhwani."I learned on the job for eight years."

His two biggest mistakes, he says now, were his failure to hire experienced management and his lack of sufficient capital. One venture-capital fund that turned down his request for money was Venrock. "Energy control companies have to deal with buildings, buildings have to deal with contractors, contractors have to deal with unions, and they all have to deal with the housing market," says Anthony Sun, who adds, in understatement: "We never felt good about that whole area."

Wadhwani sold Compuguard in 1980 to BBC Brown Boveri & Co., the Swiss maker of electrical equipment. After staying on for another year at the company, which he says was "slightly profitable," he left with the idea of buying another company. When Venrock called, proposing that he consider taking over American Robot, he came to see Anthony Sun and his colleagues in Manhattan. Perhaps only in America would an immigrant from India be recruited by an immigrant from Thailand (Sun) to use Rockefeller money in a robot company start-up.

After his meeting in Manhattan, Wadhwani met the four-man engineering team in Winston-Salem. Then he returned to Pittsburgh to study the robotics business and decide if he wanted to be part of it. "My first inclination was to run like hell from the industry," he says. "I went down several times to Winston-Salem to convince myself I wasn't crazy," Nevertheless, attracted by the professional challenge, he finally stepped into the job of president in January 1982. Wadhwani and Venrock each invested $75,000 in the company to keep it going while they tried to raise additional capital.

His first decision was to more American Robot to Pittsburgh, close to the prestigious Robotics Institute at Carnegie-Mellon and to potential customers in heavy industry throughout the so-called Rustbelt.Moving the company, though, was almost its undoing. To overcome Pittsburgh's reputation as a gritty industrial town, Wadhwani hosted his four engineers and their families on a tour of the city one weekend. "There was 18 inches of snow that weekend, and low temperatures," says Wadhwani. "It was Antarctica. By Sunday I thought I'd be starting from scratch with a new team."

The team stayed on despite the snow, however, and Wadhwani dug in. Right away he sensed that competing directly robot arms would be a kamikaze attack. The new challenge, defined by a few robot companies like Automatix Inc., of Billerica, Mass., was to marry the robot to other computerized equipment to create a complete production system. Wadhwani was not just the talent agent, providing one actress for the play; now he would supply the cast, the script, and the scenery as well.

Before developing a product line, Wadhwani had to understand the needs and problems of manufacturers that were trying to move their factories toward computer-integrated manufacturing (CIM) -- the "factor of the future" -- in which computers control every major aspect of design, engineering, and production.

The incubation of a new product at many modern facilities, from auto plants to architects' offices, takes place at a computer-aided design terminal rather than at a drafting board. Designers draw a model of an object, be it an automobile axle or an airplane wing, and then run simulated stress and mechanical tests on it. When the design is complete, the computer converts the drawing into a mathematical model and stores it in memory. Later, factory managers receive a numerical tape that instructs machine tools how to produce the object.

On the factory floor, a hierarchy of computers runs the show. A computerized controller operates each machine in a work cell, including robots, conveyor belts, and artificial-vision systems. A minicomputer coordinates a few work cells, and a mainframe computer commands all the work cells in the factory. If all the various computers can exchange information, a problem at any point in the process is communicated down the line in a fraction of a second, and the control computer orders the proper adjustments throughout the entire system.

CIM has enormous advantages over a traditional factory, beyond the obvious productivity gains. It provides unprecedented flexibility, enabling companies to make different products simply by reprogramming the computers. By eliminating or greatly reducing the need for retooling, it makes small batches of unusual items almost as cheap to manufacture as mass-produced items.

This promised land for manufacturing has arrived for only a few factories, however. Many others are handcuffed by the limitations of most robots -- which, although they enjoy a reputation for brilliance, actually have no more smarts than a head of lettuce. They are unable to adapt to the unexpected: A welding robot, for example, will mindlessly solder a door permanently shut if the assembly line goes awry and delivers a car out of position.

Even the new generation of smarter robots has a serious problem. Most robots and computers on the factory floor wouldn't talk to one another if their electricity depended on it. Made by different manufacturers, the robots use proprietary electronics that prevent them from exchanging information without the addition of costly new devices.

At General Motors Corp.'s plant in Lake Orion, Mich., for example, managers have to deal with computerized equipment from a dozen different manufacturers, each speaking a different language. "It's like putting a Spaniard, a Frenchman, a Russian, and an Italian in one room all speaking their own language and asking them to solve the world's problems," says David Burror, chief executive of Logitek Inc., which integrates computer systems in auto plants.

It was into this chaos of bits and bytes that Wadhwani brought American Robot. "We wanted to develop systems that allowed our customers to grow into an automated CIM factory step-by-step," he says. Wadhwani figured he could reach that goal with a three-part strategy. First, the company would build a robot much more intelligent -- albeit more expensive -- than most Japanese machines. Second, it would develop software to enable the robots to interact with other computerized equipment in the factory. Finally, it would team the robot with the additional equipment needed to perform an entire production process.

His strategic plan convinced a number of venture capitalists to part with their millions. In April 1982, three months after Wadhwahni took command, Koppers Corp. invested $1.6 million in American Robot. Three more investors added $2.5 million in August. In all, Wadhwani says he raised $16 million to back the company, about three-quarters of which remains unspent.

As the new money flowed in, Wadhwani began implementing his strategy. He pushed on with the development of Merlin, despite the growing movement among his competitors to buy inexpensive robots from Japan and then add their own software and peripherals. "A company that buys robots overseas can add only so much with its own software," he says. "We get maximum capability -- a powerful software controller that's not limited by the arm."

Wadhwani's engineers developed a sophisticated robot that can handle a wide range of applications in the factory with a change of computer programs, compared with the generally single-purpose arms made overseas. That versatility saved American Robot the enormous cost of developing a broad product line to handle different tasks.

Merlin has some features designed to help ease it into the factory of the future. Rather than devise yet another proprietary operating system, the internal program that tells a computer how to do its work, American Robot licensed American Telephone & Telegraph's UNIX system and then adapted it for Merlin. Many analysts think that UNIX will become the industry standard for minicomputers; if that happens, software written for Merlin will run on many other industrial computers. UNIX also enables computers to perform several tasks simultaneously, making it easier to put additional robots and automated systems on the factory floor under the command of a single computer system.

The company similarly declined to write a proprietary programming language, deciding instead to adapt Basic for its robots. Already in widespread use -- schoolchildren learn it as one of their first programming languages -- Basic could become another CIM standard. Almost any engineer familiar with computers can program Merlin, a distinct advantage in the many factories that have a smorgasbord of computer languages. "Most factory people aren't computer jocks," says Wadhwani. "We thought Basic would be a minimum threat to people."

The computers that command Merlin have a hierarchical structure patterned after corporate organizational charts. At the lowest level, comparable to that of a production-line worker, is the company's "universal controller," a powerful 32-bit computer. Its software tells a single robot arm and vision system what to do, and enables technicians to program both in the same language, Basic. The computer at level two, called a "cluster controller," has software that coordinates many lower-level controllers in a work cell or up to 64 robots performing different tasks.

The minicomputer at level three, made by another vendor, in turn will run many cluster controllers and work cells. Like a factory manager, it will issue orders to the entire facility and retrieve production data in return. At level four -- that of the chief operating officer -- a big mainframe computer will coordinate production scheduling, inventory management, and quality-control reporting.

In addition to the robots and the software, Wadhwani's third major focus is on supplying complete systems for production tasks. American Robot introduced a conveyor belt last June that comes in two-foot-long modules; customers can fit together up to 100 of the modules to construct exactly the product they need. The company is also selling an artificial-vision system that gauges or inspects parts and assemblies.On the factory floor, the same computer and software that control Merlin can control the conveyor belt and vision system, too.

American Robot hasn't been around long enough to have much of a track record, but several major companies are using or testing its robot. At Storage Technology Corp., four Merlins carry out some of the critical steps in the assembly of disk drives. Because the smallest specks of dust can ruin a disk drive, scrap rates are very high even in the best "clean rooms" where they are made.

The Merlins take over some of the handling tasks normally performed by human operators, who cause much of the contamination problem. Two robots help assemble the disk drives, working with a precision that people can't attain. Mounted on 25-foot-long tracks, two other Merlins carry the disk drives to the testing stations. All four communicate with the other computerized equipment in the room.

Another electronics company, Fairchild Camera and Instrument Corp., is using Merlin to test printed circuit boards. Merlin picks a circuit board from a bin and position it on an automatic tester, which tells it a few seconds later whether to place the board in the active bin or the repair bin. American Robot is pursuing other applications as well: In welding, for example, a company can buy equipment from outside vendors and integrate it with Merlin; then it can sell the complete system, ready to use.

Because developing and selling these systems is so costly, the lack of strict controls can land even the best technology in the hands of a bankruptcy judge. To hold down expenses, Wadhwani contracts out production of the Merlins. He is also trying to avoid developing one-of-a-kind systems; instead, he emphasizes such systems as the welding robot, which the company could sell to hundreds of customers. Finally, by licensing the technology to foreign companies -- American Robot has agreements in the United Kingdom as well as in Japan -- he can increase sales without additional marketing expenses.

Wadhwani will have to circle on the mat for some time before it becomes clear whether American Robot can survive as one of the profitable few. Even if his strategy proves wise, other companies may implement it better. It seems certain, though, that the company will win or lose under the new rules of competition -- as a second-generation manufacturer selling systems and software instead of just robots.

In this next round, with American software married in many cases to cheap Japanese robots, U.S. companies could win back the market they lost decisively to Japan in the 1970s. In the process, they seem likely to reshape American factories as well.