Fuzzy Visual Control for Memory-Based Navigation Using the Trifocal Tensor

In this paper, a fuzzy control scheme for visual path-following of wheeled mobile robots is presented. It relies on the feedback of a geometric constraint: the trifocal tensor (TT). The TT is computed from the image currently seen by an onboard camera and the sequence of target images previously acquired, which defines the visual path. Only one element of the TT is needed for feedback, which provides information of the robot’s deviation from the path. This is used in a set of Mamdani-type fuzzy rules that mimics the human action of driving a vehicle. The translational velocity is also adapted by a fuzzy system in function of the path’s curvature, inferred from the TT computed a priori using the set of target images. The use of fuzzy control allows achieving an effective and simple controller that does not need a time-varying reference to be tracked while the resultant robot velocities are smooth or piece-wise constant. The validity and performance of the approach is shown through simulations using synthetic images.