Behaviour Cooperation by Negation for Mobile Robots

This article presents a behavioural architecture, the Survival Kit (SK), which allows behaviours to cast their multivalued output by means of constraints over an ´action feature space´. In other multi-valued based behavioural architectures behaviours cast their output directly over a set of ´action spaces´. As the number of degrees of freedom increases, the amount of data flowing from behaviours to the coordination node, the required memory, and the computational demand to select the best action increase exponentially. By using an ´action feature space´, the SK reduces greatly these problems. Furthermore, the ´action feature space´ is endowed with inertia, which provides the architecture with a mechanism to automatically handle noisy sensors, to create fixed action patterns, and to maintain local implicit representations of the environment. Special attention is given to an environmental feature, the gap, which allows to optimise paths based on immediate ranging data. Experimental results in a physical robot demonstrate the ability of the architecture to control robots with noisy sensors.