Reliable Long-Term Navigation in Indoor Environments

Abstract

Long-term robustness is a crucial property of robots intended for real-world tasks such as, for example, transportation in indoor environments (e.g. warehouses, industries, hospitals, airports etc.). In order to be useful, such robots must be able to operate over long distances without much human supervision, something that sets high demands both on the actual robot (hardware) and its artificial brain1 (software). This paper is focused on the development of artificial robotic brains for reliable long-term navigation (for use in, for example, transportation) in indoor environments. Reliable decision-making is an essential tool for achieving long-term robustness. The ability to make correct decisions, in real-time and often based on incomplete and noisy information, is important not only for mobile robots but also for animals, including humans (McFarland, 1998; Prescott et al., 2007). One may argue that the entire sub-field of behavior-based robotics emerged, at least in part, as a result of the perceived failure of classical artificial intelligence to address real-time decision-making based on a robot’s imperfect knowledge of the world. Starting with the subsumption method (Brooks, 1986), many different methods for decision-making, often referred to as methods for behavior selection or action selection, have been suggested in the literature on behavior-based robotics (see, for example, Bryson (2007); Pirjanian (1999) for reviews of such methods). In actual applications, a common approach is to combine a reactive layer of decision-making, using mainly behavior-based concepts, with a deliberative layer using, for example, concepts from classical artificial intelligence, such as high-level reasoning) (Arkin, 1998). Several approaches of this kind have been suggested (see, for example, Arkin (1987) and Gat (1991)) and applied in different robots (see, for example, Sakagami et al. (2002)). In the utility function (UF)method for decision-making (Wahde, 2003; 2009), whichwill be used here, a somewhat different approach is taken in which, for the purposes of decision-making, no distinction is made between reactive and deliberative aspects of the robotic brain. In this method, an artificial robotic brain is built from a repertoire of brain processes as well as a single decision-making system responsible for activating and de-activating brain processes