Silva-2012a

Book chapter

In the context of the RoboCup Middle Size League, as well as any other multi-agent application, the merging of information from several sources, from the same or different agents, in an organized and useful data bank is one of the most important tasks. Moreover, a robot usually has multiple sensors and sources of information providing data of varied nature. It usually also has a decision process which decides what to do and how to react to the current conditions of the environment surrounding it. It is between these two layers that a sensor integration process is applied, to create the representation of the world, named world model. In robotic soccer, the treatment of information from obstacles on the field is one of the integration tasks. This chapter gives an insight of general steps on the creation of good obstacle representation in the robot world model. The first step is the detection of the obstacles in the image acquired by the camera of the robot. Using radial search lines and color based blob detection, the obstacles can be differentiated in the image. After visually detecting the obstacles, this information is merged with a-priori known information about the obstacles, such as their maximum size, the position of the robot on the field, among other information. This allows some discrimination between real obstacles and possible false positives. Using the capability of sharing information in this multi-agent application through the use of an advanced Real-Time Database and a communication system, the information from the other robots can be used to improve the obstacles information of each robot. After the filtering process, the information about the detected obstacles is merged with the position of the teammate robots allowing to discriminate which of the obstacles are team robots and which are opponent robots. Finally, and with the purpose of having a team world model as coherent as possible, the robots share the obstacle information with each other. The work presented in this document was developed and tested in the CAMBADA robotic soccer team, as the means to provide the agent with more relevant information, so it can act differently when faced with team or opponent robots.