No, sorry, not the secret kind. Although if you don’t know about software agents, they may seem secret and mysterious. And I’m sure 007 would have found a use for them.
Learn about agents and one of their many uses in Cutter Consortium Senior Consultant James Odell’s recent E-Mail Advisor. It’s reproduced below, and your comments are most welcome.
Agent Technology: Painting Trucks at General Motors
Traditionally, assembly line schedules are centrally developed and controlled. Any change in the schedule must be centrally reconfigured. When the line is small and has few unplanned stoppages, centrally controlled schedules work well. However, scheduling for most real-world assembly lines can be a nightmare: work stations break down, personnel get sick, environmental conditions are not always within acceptable limits, products coming down the line have or acquire unexpected defects, and so on.
Dick Morley, a technology visionary and father of the programmable controller, swept away old assembly line schedules and developed a better system for painting trucks at GM’s assembly plant in Fort Wayne, Indiana. “How do I schedule the nonschedulable?” Morley wondered. “Trucks do not come down the line in order of their color, and frequently no paint booth is available with the correct color.” Morley also discovered that many of the paint booths were typically broken down or being repaired.
In his technique, the scheduling program interacts with each paint booth. Instead of assigning unpainted trucks to booths, GM’s solution was to have the booths bid on the paint jobs. To accomplish this, each booth was equipped with a simple software agent that was programmed to keep its booth busy and bid on each paint job. The amount of the booth’s bid was based on how busy the booth was at the moment of bidding, whether it had to change to a different paint, and whether the booth was functioning properly.
To coordinate the various bids for each paint job, a scheduler agent acts as a broker. For example, when a truck arrives to be painted, the scheduler agent tells the booths, “I have a truck that needs to be painted red.” A vacant paint booth already loaded with red paint will bid very high. However, a vacant booth with a different color would bid lower because of the extra labor and time to clean and reload the paint gun. A booth that has just started to paint a truck, has broken down, or is otherwise less suited for the job would bid even lower. Based on the outcome of the bidding activity, the scheduler assigns the truck to the highest-bidding paint booth.
In a top-down, planned, “push-through” world, if one booth malfunctioned, a centrally controlled system would require immediate recomputing. With bottom-up, “pull-through” paint booth agents, other booths were ready to pick up the bidding slack at a moment’s notice. This new design saved $1 million in nine months and reduced the lines of computer code from hundreds to four. Morley and GM tackled a problem where centralized scheduling did not work efficiently by adopting an agent-based approach where each booth acts on its own behalf using a market-based bidding system. Even though the scheduler was a centralized element, it deferred to distributed booth agents. Agent-based solutions do not remove centralization; instead, they try to balance it with distributed solutions wherever it makes sense.