Researchers Say Teamwork, Coordination Could Improve Robots' Soccer Skills
In an attempt to design better robot soccer players, researchers are investigating how increasing the coordination among the robot players on a five-player team could improvement the overall performance of these artificial intelligence machines. The scientists' strategy involves reinforcement processes, where coordination among robots is learned throughout the course of the game.Researchers Kao-Shing Hwang, Yu-Jen Chen, and Ching-Huang Lee from the National Chung Chen University in Taiwan have published their study on robotic soccer coordination in a recent issue of IEEE Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans. They suggest that the reinforcement learning strategy could be applied to the Federation of International Robot soccer Association (FIRA) and the Robot World Cup Soccer Games and Conferences, among other multi-agent environments.
One of the key benefits of reinforcement learning is its simplicity: the strategy uses only reward and punishment in achieving a specific task, such as occupying a position or scoring a goal. Reinforcement signals direct the robot's behavior so that it tends to maximize the long-term sum of values of the reinforcement signals. Over time, an artificial intelligence machine can learn to make the best decision by repetitive trial and error.
The researchers emphasize the importance of using different strategies within the reinforcement system, such as offensive and defensive, according to the situation in the game. The group demonstrated that using multiple strategies is better than using a single strategy, partly because a multi-strategy method can work well on both offense and defense, and other special situations. The strategies are determined by weighted roles, and when the weights of the roles change, a different strategy is employed.
In a soccer game, the main task of each robot is to occupy an appropriate position, and "work together" to move the ball forward. In doing so, the robots must take into account information on the game field, such as the boundaries of the game field, as well as the ball's position, the robots’ positions and angles, and the opponents' positions.
Because the game is complex in this way, it requires high-level strategy learning rather than basic behavior learning. In Hwang, Chen, and Lee's strategy, the system architecture is hierarchical, so that each part can be improved independently. Of course, designing a team of robots that can be coordinated to play soccer efficiently could also lead to more socially significant issues, such as search and rescue missions, and other applications.
Source: Robot World News