Search

Navigation in large-scale environments is composed of different local tasks. To achieve smooth switching between these tasks and thus a continuous control signal, usually a precise map of the environment and an exact pose estimate of the robot are needed. Both are hard to fulfil for experiments in real-world settings. We present a system that shows how one can relax the need for accurate metric models of the environment while at the same time achieving smooth task switching. To facilitate this scheme the dynamical systems approach is used, which incorporates behaviour coordination through competition in a dynamic framework. Feature detectors use sonar data to provide means for local navigation. This ability combined with a simple topological map constitutes a complete navigation system for large-scale office environments. Experiments showed that a Scout robot using this scheme is able to successfully navigate through our whole institute. Through the use of the dynamic behaviour coordination, switching between the navigational tasks occurs in a smooth manner leading to continuous control of the platform.

Explanations of cognitive processes provided by traditional artificial intelligence were based on the notion of the knowledge level. This perspective has been challenged by new AI that proposes an approach based on embodied systems that interact with the real-world. We demonstrate that these two views can be unified. Our argument is based on the assumption that knowledge level explanations can be defined in the context of Bayesian theory while the goals of new AI are captured by using a well established robot based model of learning and problem solving, called Distributed Adaptive Control (DAC). In our analysis we consider random foraging and we prove that minor modifications of the DAC architecture renders a model that is equivalent to a Bayesian analysis of this task. Subsequently, we compare this enhanced, “rational,” model to its “non-rational” predecessor and a further control condition using both simulated and real robots, in a variety of environments. Our results show that the changes made to the DAC architecture, in order to unify the perspectives of old and new AI, also lead to a significant improvement in random foraging.

An autonomous robot navigating in its environment needs a map representing the large-scale structure of the setting. There are a variety of different types of maps used in mobile robotics: for example, grid maps, maps containing geometrical beacons, or topological maps. Each of those poses its own constraints on the problem of acquiring the map, where its complexity is a major issue. We present a system that is able to build a topological map of large-scale indoor environments in a semi-autonomous way. This means that a person shows the robot around, interacts with it via a basic interface, the platform is following autonomously, and simultaneously creating the map. The acquisition mechanisms are rather simple and computationally inexpensive, due to the low complexity of the topological map. However, experiments show that this map is sufficient to navigate between arbitrary positions in the environment.

Behaviour coordination is a notorious problem in mobile robotics. Behaviours are either in competition or collaborating to achieve the goals of a system, which leads to requirements for arbitration and/or fusion of control signals. In most systems the arbitration is specified in terms of "events" that denote positions or sensory input. The detection of these events allows discrete switching between groups of behaviours. In contrast, the fusion of behaviours is often achieved using potential fields, fuzzy rules, or superposition. In most cases, the underlying theoretical foundation is rather weak and the behaviour switching results in discrete changes in the overall system dynamics. In this paper, we present a scheme for behaviour coordination that is grounded in the dynamical systems approach. The methodology provides a solid theoretical basis for analysis and design of a behaviour coordination framework. This framework is demonstrated in the context of a domestic robot for fetch-and-carry type tasks. It is here shown that behaviour coordination can be analyzed as an integral part of the design to facilitate smooth transition and fusion between behaviours.

The dynamical system approach to behaviour based robotics provides a sound mathematical framework to behaviour design and integration. So far this approach has been used in various simulation work and in simple real world settings. This paper is a first step towards application in a more realistic scenario such as a typical office environment. The robot perceives its environment via a set of sonar sensors. Simple geometric representations of the corridor and obstacles are extracted from the perceptual information. Obstacle avoidance and corridor following are implemented as the basic behaviours, which allow the robot to navigate safely through its environment. The robot can reliably distinguish between passages that are either wide enough to pass or too narrow to be traversed. Coordination among the primitive behaviours allows the robot to cope with more complex situations, such as a corridor that is blocked by obstacles, in a flexible manner.