ARUP; ISBN: 978-0-9562121-5-3 - CMBBE 2012 - Cardiff University

NEURAL NETWORK-BASED PREDICTION OF HAND POSTURES
FOR GRASPING SIMULATION WITH A BIOMECHANICAL MODEL
M.C. Mora 1 , J.L. Sancho-Bru 2 , J.L. Iserte-Vilar 1 and A. Pérez-González 3
1. ABSTRACT
Most human interactions with the environment are performed by the hands. This
versatility is due to their more than 20 degrees of freedom. A lot of research has been
done in the field of robotic grasping in the last years [1,2]. All this work can be
extended to the biomechanics field due to the similarity between the grasping problem
in robotic and human hands, considering that the former are kinematically simpler than
the latter. In [3] we proposed a hand biomechanical model that merged the current
knowledge on biomechanics, ergonomics and robotics. One of the aims of this proposal
was to predict feasible grasping postures for a given object. As a first approach, we
present here a methodology for automatically generating grasping postures by means of
artificial neural networks (ANNs), inspired by the works presented in [1] and [4]. The
inputs to the ANN are the characteristic hand data, the dimensions of the object to be
grasped and the task. The outputs are the joint angles of the hand. For the training,
validation and test stages, we use data measured from a Cyberglove device. Our goal
is to estimate the hand posture performed when grasping objects of the daily life.
2. INTRODUCTION
The study of the human grasp is very interesting because of the knowledge it can
provide regarding manipulation [5]. In the last two decades a lot of research has been
carried out in the field of robotic grasping [1, 2, 5, 7] and is currently a very hot topic.
Many of the existing techniques could be extended to the biomechanics’ field due to the
similarities between human and robotic hands, although the former are simpler that the
latter. Within the wide variety of grasp-related problems, the computation of grasping
postures associated to particular objects is one of the most challenging, as it implies the
fulfillment of a large number of constraints that relate not only to the hand structure and
to the object, but to the requirements of the environment and the task to be performed.
From the biomechanics’ viewpoint, there are different studies of the grasping process
but very few tackle the prediction of grasping postures for given objects and tasks. Hand
posture prediction is an important issue as it allows the evaluation of biomechanical and
ergonomic parameters related to grasping [7, 8]. However, the few related works
usually measure hand postures using specific devices, such as data gloves [9], instead of
1
Assistant Professor, Mechanical Engineering and Construction Department, Universitat Jaume I, Avda.
Vicent sos Baynat, 12071, Castellón, Spain
2
Associate Professor, Mechanical Engineering and Construction Department, Universitat Jaume I, Avda.
Vicent sos Baynat, 12071, Castellón, Spain
3
Professor, Head of the Mechanical Engineering and Construction Department, Universitat Jaume I,
Avda. Vicent sos Baynat, 12071, Castellón, Spain