Behavior Fusion for Visually-Guided Service Robots

Abstract

Mobile service robots are the class of robots that have tools to understand the environments at home and office. The development of mobile robots is increasing world-wide due to the availability of moderate price sensing and computing devices. Moreover, there is a strong belief that the market for service robots is just about to undergo a radical increase in the next few years. Despite the huge literature of the mobile robot navigation, the development of intelligent robots able to navigate in unknown and dynamic environment is still a challenging task (Walther et al., 2003). Therefore, developing techniques for robust navigation of mobile robots is both important and needed. The classical approach for mobile robot control used the Model, Sense, Plan, Act, MSPA serial strategy, which proved to be inherently slow and totally fails if one module is out of order. We may call this approach as a planner-based control approach. The appearance of behavior-based navigation approach (Arkin, 1998; Brooks, 1986) was a remarkable evolution, in which the reactive behaviors were designed to run simultaneously in parallel giving tight interaction between sensors and actuators. The reactive behaviors allow for incremental improvements and addition of more application-specific behaviors. Building several behaviors, each concerned with a sole objective, will produce different decisions for the robot control parameters, and they have to be combined in some way to reach the final motion decision. The fusion of independent behaviors is not an easy task and several approaches were proposed in the literature to solve this problem (Arkin, 1998; Borenstein & Koren, 1991; Carreras et al., 2001; Saffiotti, 1997). Coordination of behaviors can be classified into two further approaches, a competitive, as was originally proposed by (Brooks, 1986), and cooperative strategies (Carreras et al., 2001). Depending on the environment, the competitive approach may fail and become unstable in critical situations demanding higher switching frequencies between behaviors. In the subsumption architecture (Brooks, 1986) behaviors are activated once at a time but this may be inadequate for a variety of situations requiring several behaviors to be active at the same time. In the cooperative approach, all behaviors contribute to the output, rather than a single behavior dominates after passing an objective criterion. An example of the cooperative O pe n A cc es s D at ab as e w w w .ite ch on lin e. co m