Hierarchical grasp controller using tactile feedback

Grasp stability is a challenging problem in robotics. It needs to be robust to external perturbations and adapt to unknown objects. While performing a stable grasp, grip strength control can be a desirable property for many applications. We present an approach for stable object grasping and simultaneous grip strength control using tactile feedback, which is able to deal with unknown objects of different shape, size and material. We develop a generic method that exploits the structure of an anthropomorphic hand to be simple and effective. Our approach uses techniques from classical control theory to develop a controller in charge of coordinating the fingers for achieving grasp stabilization and grip strength control. Then, we applied a machine learning method based on Gaussian mixture model regression, with the aim of further improving stability and increasing robustness to external perturbations. The method has been validated on the iCub robot. Experimental results show the effectiveness of our approach.