Design of a Neural Network Power Grasp Controller

by Mark D. Hanes, Ohio State Univ, Columbus, United States,
Stanley C. Ahalt, Ohio State Univ, Columbus, United States,
David E. Orin, Ohio State Univ, Columbus, United States,



Document Type: Proceeding Paper

Part of: Mechanics Computing in 1990's and Beyond

Abstract:

An overview of research on the development of an artificial neural network controller used to implement power grasping on the DIGITS Grasping System is presented. The DIGITS System, under development at The Ohio State University, consists of four, three-degree-of-freedom fingers that can be used for grasping and dexterous manipulation. This paper presents two techniques, adaptive activation-function slopes and modular networks, which reduce the required training time and yield comparable or improved performance with respect to traditional fixed-sloped networks. A comparison of the techniques based on the desired force distribution power grasp control is also discussed.

Subject Headings: Neural networks | Degrees of freedom | Control systems | Training | Slopes | Research and development | Power transmission

Services: Buy this book/Buy this article

 

Return to search