AUSTIN, Texas—A team of computer scientists from The University of Texas at Austin has won an international contest where computer-simulated agents were pitted against each other in a virtual manufacturing game.
Dr. Peter Stone and his team designed the winning agent, named TacTex-05, an autonomous computer program that can learn, experiment and behave independently of human control.
In the growing global economy, autonomous agents like TacTex-05 could have tremendous impacts on supply chain manufacturing processes, like assembly of automobiles and computers. Most supply chains are managed by humans. They are responsible for coordinating multiple manufacturers and parts suppliers and making speculative decisions based on quickly changing markets.
“Autonomous agents may be able to make much quicker, more flexible decisions,” says Stone, assistant professor of computer sciences. “This would make the entire supply chain more efficient and responsive to environmental changes.”
In this year’s Trading Agent Competition (TAC) finals, held from Aug. 1-3 at the International Joint Conference on Artificial Intelligence (IJCAI) in Edinburgh, Scotland, the agents competed against each other in a game simulating the manufacture of personal computers.
The agents managed parts inventories, competed for customers, negotiated with suppliers and performed a range of other decision-making tasks over a virtual 220-day period. The agent who finished the game with the most money in the bank was declared winner.
“We believe that TacTex’s main advantage was its ability to procure supplies more cheaply than other agents, and to target the most profitable markets,” says Stone. “Our agent was also one of the only ones that effectively adapted its strategy from game to game based on the results of previous games.”
TacTex-05 lives on a computer at The University of Texas at Austin, and it competed with the other agents networked to the game running on a machine in Sweden. Thirty-two teams from universities and companies around the world challenged each other for the top spot during the entire TAC. In the final round, TacTex-05 averaged more than $4.7 million in profits per game, far more than any other agent.
Stone says the yearly competition’s main participants are from the academic community. He says many of the agents in next year’s competition will be strongly influenced by the successes of TacTex-05. In this way, computer scientists will continue to refine the adaptability and decision-making processes of autonomous agents, with the goal of increasing efficiencies in supply chain management using smarter computers. Eventually, the methods developed for TAC agents such as TacTex-05 may find applicability in real supply chains.
The TacTex-05 team included Stone, graduate student David Pardoe, undergraduate Jan Ulrich, and high school student Mark VanMiddlesworth from St. Stephen’s Episcopal School in Austin.
For more information contact: Lee Clippard, College of Natural Sciences, 512-232-0675.
