Optimal object grasping using fuzzy logic

摘要

Robotic end effectors are commonly required to hold and manipulate various objects under a wide range of conditions.To achieve a satisfactory grip, optimal force control is required to avoid the risk of the object slipping out of the end effector and to avoid any possible damage to the object.In this paper, we investigate the use of fuzzy logic controllers to achieve optimal grasping.Starting with a set of manually-designed fuzzy rules from previous work, these were analysed and a shortcoming identified and corrected.Backpropagation learning was then employed to train a neurofuzzy controller with and without a priori knowledge.In the former case, the a priori knowledge which defined the start point for learning was obtained from the manually-designed rules.In the latter case, training started from different initial random weight settings.The learned solutions were very similar although training was faster with a priori knowledge.The learned neurofuzzy controller had better performance than the manually-designed fuzzy controller, particularly in respect of its faster control action.