Scalar sensor-based adaptive manipulation for source seeking

Abstract

This paper considers a class of robotic manipulation that can automatically trace an unknown source of a scalar field by sensors attached to links of a robotic manipulator. To achieve this, one approach is to model the field map by a radial basis function (RBF) network and to update its weights in a recursive way so that the gradient estimate can be available in realtime to command the end-effector toward the target source. In this paper, we investigate the practical implementation of this autonomous manipulation scheme and demonstrate its performance through experimental tests. Firstly, we provide a selection guideline for the Gaussian-type RBF network. Secondly, the field estimation algorithm is simplified to a suboptimal estimator instead of the original recursive least square (RLS) filter previously used. Thirdly, a cross-coupled parameter estimator is newly introduced for global convergence of the combined control law. The overall control scheme is experimentally demonstrated using a two link planar robot. A smooth gray scale map is devised to represent the unknown physical potential field with its scalar values measured by color sensors on robot links. The effect of under fitting of the field model is also investigated through the experimental results.