Multiple Omnidirectional Vision System and Multilayered Fuzzy Behavior Control for Autonomous Mobile Robot

Abstract

In the research on multiple autonomous mobile robots such as RoboCup, some methods for obtaining the environmental information over all circumferences used an omnidirectional vision sensor, were proposed. In case of the research using the omnidirectional camera (called omni-camera), only one camera is almost used in general. However, the object image in the mirror is compressed according to the distance. If the height of the object is uncertain, the accurate distance measurement is generally impossible. To solve these problems, some researches for the stereo vision system used two omnicameras were also proposed. For example, the research for the stereo vision system which two omni-cameras are vertically fixed was proposed by J.Gluckman (Gluckman; Nayar & Thoresz, 1998), H.Koyasu (Koyasu; Miura & Shirai, 2002) and T.Matsuoka (Matsuoka; Motomura & Hasegawa, 2003). The other approach which two omni-cameras are horizontally fixed is proposed by R.Miki (Miki et al., 1999). In our laboratory, we have developed a multiple omnidirectional vision system (called MOVIS) which three omnidirectional cameras are arranged on an autonomous soccer robot like as a horizontal and equilateral triangle (Shimizuhira & Maeda, 2003). As a result, the stereo-vision system by the principle of the triangulation is made by each two cameras. The purpose of this research is to realize the object recognition and the position measurement of the robot accurately in real time. Furthermore, we propose the real-time object position measurement and the self-localization method for the autonomous soccer robot with MOVIS. On the other hand, there are some researches for the autonomous behavior under the complicated environment by using fuzzy reasoning. In the research of the behavior control in the RoboCup middle-size league, a control system based on the fuzzy potential method was proposed by R.Tsuzaki (Tsuzaki & Yosida, 2003), a multi-layered learning system was proposed by Y.Takahashi (Takahashi; Hikita & Asada, 2003). Generally, it is well known that an operator is easy to express his control knowledge by using fuzzy reasoning. We have already proposed a multi-layered fuzzy behavior control method that element behaviors of the robot are individually controlled with the behavior decision fuzzy rule in lower-layer, and combined them with the behavior selection fuzzy rule in higher-layer (Shimizuhira; Fujii & Maeda, 2004) (Maeda & Shimizuhira, 2005).