Neural-network based planner of force for multifingered grasp

The objective of this paper is to develop a general purpose planner of grasping force which will allow multifingered hands to firmly grasp an object of arbitrary shape. An artificial neural network (ANN) based planner of force for precision grasping is discussed. A nonlinear programming method is applied to train the ANN. In the method, the objective function is the minimum norm contact force. Friction constraints, force constraints and joint torque constraints are considered. The ANN used for this research is based on the functional link (FL) network without hidden layers. The ANN can be trained by an example. 121 samples across the usable input space are included in the training set. The results obtained by simulation show that the ANN is able to learn the appropriate nonlinear mapping between the object size and joint torques to an acceptable level of accuracy and can be used as a real time precision grasp planner