ASU team wins AI commentator track of the KAIST World Cup 2018
This summer, Arizona State University organized an interdisciplinary team for the AI commentator track of the Korea Advanced Institute of Science and Technology (KAIST) AI World Cup 2018, a worldwide competition of online soccer teams simulated by AI. AI commentators give play-by-play commentary and short-term predictions on AI soccer matches.
Assistant Professor Yezhou Yang in the Ira A. Fulton Schools of Engineering led the team comprised of Chia-Yu Hsu, a master’s student in computer science, and Siyu Zhou, a doctoral student in physics. Hsu and Zhou spent a summer working in Yang’s ASU Active Perception Group’s vision and language special interest sub-group.
Yang’s group focuses on cognitive robotics, computer vision and robot vision to program intelligent agents. The group seeks solutions to visual learning that involve deep learning and AI algorithms to enable agents to interpret people's actions and a scene’s geometry.
The team programmed rules for the commentator agent based on a human’s observation of the playfield. Basically, a human commentator can look at the current states of motion of the players while being aware of those in the past. As such, the team submitted a learning-based model that enables the agent to interpret the current situation of a robot soccer game similar to a human commentator by observing past actions regarding the soccer ball’s location and speed.
It took about two weeks to write the comment-generating scripts, and another week to enrich the comments and rules.
Hsu and Zhou won the AI commentator competition and received a $5,000 prize for the quality of their play-by-play commentaries and short-term predictions of two soccer matches designated by an evaluation committee.