Action decision with incomplete information: a temporal database approach

The agent-oriented paradigm is widely used for developing flexible information systems. Intelligent agents such as soccer agents decide autonomously their actions according to the information they can obtain. The more information they can get, the more appropriate action they can decide. However, the agents usually run in a real-time environment, and they can not obtain the complete information that they require. For example, a soccer player may lose and quantize some visible data about the placement of either the players or the ball. In this case, the player has to decide its behaviors according to incomplete and even unreliable information. Hence, it is an important for intelligent agents to complete some lost information according to the history of their behaviors. In the paper, we design a temporal database for the soccer agents to record the history of the behaviors of all objects. While the new information comes, the database can complete and correct them by itself according to the temporal information stored in the database. For example, when a soccer agent loses the exact location of the ball at time t, the agent can infer from the location of the ball and the closest player at time t-1 that are recorded in the database. This way, the agents can behave more exactly than those without the temporal information. Furthermore, we implemented the soccer agents with the database, and experiments showed that the database can complete information for the soccer agents in real time.