Vision-based fuzzy controllers for navigation tasks

This paper deals with vision-based fuzzy closed-loop control schemes for collision avoidance as well as maintenance of clearance in a-priori unknown textured environments. These control schemes employ a visual motion cue, we call the visual threat cue (VTC) that provides some measure for a relative change in range as well as clearance between 3D surface and a fixated observer in motion. It is a collective measure obtained directly from the raw data of gray level images, is independent of the type of 3D surface texture. This motion cue is scale-independent, rotation independent, needs no 3D reconstruction and is measured in [time^-1] units. Fuzzy control is closer in spirit to human thinking and can implement linguistically-expressed heuristic control policies directly without any knowledge about the dynamics of the complex process. The fuzzy controllers were implemented in real-time using a 486-based personal computer and a camera capable of undergoing 6-DOF motion. Results are highly encouraging