Object Recognition and Visual Servoing: Two Case Studies of Employing Fuzzy Techniques in Robot Vision

Summary

The capabilities of observing the world and manipulating objects based on visual information are basic requirements in robotic applications. Typically, for manipulating objects in real environments it is necessary to recognise and locate the substantial objects robustly. Due to external influences, e.g. partial occlusions of objects and illumination changes, as well as to internal influences, e.g. noisy imaging hardware, inaccurate measurements and quantization effects, the recognition systems have to cope with incomplete, uncertain and inaccurate information.

In the first part of this chapter we present the general framework of a robust approach for recognising partially occluded objects. It combines the popular concept of using invariant shape descriptions, i.e. descriptions of objects which remain unaffected by certain variations of the intrinsic and extrinsic camera parameters, with the flexibility and readability of rule-based fuzzy systems by applying invariant object shape descriptions in fuzzy if-then classification rules.

In the second part, we propose a fuzzy control approach for learning fine-positioning of parallel-jaw robot gripper using visual sensor data. The first component of the used model can be viewed as a perceptron network that projects high-dimensional input data into a low-dimensional eigenspace. The second component is a fuzzy controller serving as an interpolator whose input space is the eigenspace and whose outputs are the motion parameters. Instead of undergoing cumbersome hand-eye calibration processes, our system is trained in a supervised learning procedure using systematical perturbation motion around the optimal grasping pose.