Integrating Behaviors for Mobile Robots: an Ethological Approach

Abstract

Robots available nowadays in the everyday markets can be divided into two major groups. First, those that are oriented to a single task, as vacuum cleaners (roomba1, robocleaner2, ...), lawn mowers (automower3, robomower4, ...), etc; and second, those oriented to various tasks, as robotic pets (Aibo5, Necoro6,... ), museum guiders, or research platforms (Pioneer, Koala, etc.). In order to get service robots or personal assistants we need to improve the abilities of the second ones. These abilities have to do with their ability to smartly combine their basic skills to obtain behaviors that are more complex. The generation of autonomous behavior is a very complex issue. Within a multi-goal system, like service robots, the focus is more on the integration than on the control of perception algorithms. The organization becomes the critical issue; robustness and flexibility are key features. As animals can do it, we can see no compelling reason why robots could not, but more research on robot architectures is needed to reach an acceptable performance. Several paradigms have been historically proposed for behaviour generation in robots. These paradigms are also known as architectures. Dynamic and uncertain environments forced the evolution from symbolic AI (Nilsson 1984) to reactive and behavior based systems (Brooks 1986; Arkin 1989). The behaviors based systems adapt smoothly. They have no anticipation and no state, but they have shown poor scalability for complex systems. Hybrid architectures have been predominant since mid 90s (Simmons 1994; Konolige 1998; Bonasso 1997), mainly those three-tiered ones that add two layers to behavior based ones, usually a sequencer, and a deliberator. Several architectures have been explored after the hybrid three-tiered architectures became the de facto standard. In particular, many reviews of the hierarchy principle have been proposed in last years (Arkin 2003, Saffiotti 2003, Nicolescu 2002, Behnke 2001, Bryson 2001), trying to overcome subsumption limitations. Our contribution in this area is a novel hierarchical approach named JDE. This new hierarchical approach, ethologically inspired, is based on the selective activation of