Carnegie Mellon University’s computer poker team dealt its strongest performance to date, decisively beating its competitors in Heads-Up, No-Limit Texas Hold’em at the Annual Computer Poker Competition at the Association for the Advancement of Artificial Intelligence (AAAI) meeting in Quebec City.
Meanwhile, Carnegie Mellon’s team in the annual RoboCup robot soccer competition, the winningest team in the history of RoboCup’s small-size league, had another strong outing at the world championship in João Pessoa, Brazil, though ultimately the team came in second.
At the AAAI, the CMU team included Tuomas Sandholm, professor of computer science, and two of his Ph.D. students in computer science — Noam Brown and Sam Ganzfried. They competed against 13 other teams in the Heads-Up, No-Limit Texas Hold-em game, which Sandholm says is the game that all of the top teams focused on this year and in which the most progress has occurred in recent years.
Much as computer chess was an early milestone for artificial intelligence, computer poker has emerged as the next important challenge. In contrast to chess or Go, poker involves hidden cards and thus forces the teams to compete based on incomplete information.
The game included two categories of competition — elimination, in which the bots all play together and the weakest bot is eliminated at the end of each round, and total bankroll, where the goal is to win as much as possible from all opponents in aggregate. The CMU team, Tartanian 7, won both categories in the competition that concluded July 27.
“There were statistical ties for second, third, and fourth, but we won both categories with clear statistical significance,” Sandholm said.
At RoboCup, July 19–25, the CMDragon team included Manuela Veloso, professor of computer science, two of her Robotics Institute Ph.D. students, Joydeep Biswas and Juan Pablo Mendoza, and two Computer Science Department Ph.D. students, Danny Zhu and Richard Wang. The team won seven games to get to the final, a rematch of last year’s final with a team from China’s Zheijang University.
“The game was very good, even if we lost 2-0,” Veloso said.
Like computer poker, robot soccer is a major scientific challenge, requiring teams to create algorithms and tactics that enable their teams to work cooperatively toward a goal in an adversarial environment. Veloso, a founder of RoboCup, discussed the progress being made in this area in an interview with The Economist.
Byron Spice | 412-268-9068 | bspice [atsymbol] cs.cmu.edu