Dynamics for vision-based autonomous mobile robots

We present a dynamical systems approach to generate basic navigation behaviors implemented on an autonomous mobile vehicle. In this approach, a behavior is expressed as a stable state of a variable in a dynamical systems equation. The value of this behavioral variable is specified by incoming information which is provided by the sensory systems typically but can also be determined by other behavioral levels. Multiple sources of information act as forces in the vector field of the dynamical equation that describes a specific behavioral level. Each force votes for a certain value of the behavioral variable, it has a strength that represents the relative importance or certainty of the specified information and it is supplied with a range, that determines the overlap between the multiple incouplings. As an example for this approach on the sensor side, we describe a dynamical optical flow estimation that takes advantage of the favorable stability properties of this type of dynamical systems as robust estimators particularly suited for ambiguous and noisy visual information. On the effector side we present an obstacle avoidance scheme that employs a dynamical system for steering a visually guided mobile robot platform in an indoor environment with obstacles. By coupling of the sensor- and the effector system, a closed loop navigation behavior is achieved.