In this paper, we present a robotic grasping system for deployment in perso
nal robots. The system learns how to grasp objects from experiments. This a
pproach allows it to satisfy a number of requirements that we have identifi
ed as prerequisite for operation in personal robot environments. The system
design consists of three control layers, each describing the control strat
egy of a predefined behavior. Learning of the behavior ib performed using g
roups of neural networks. Testing of the system was performed in a simulate
d environment using a specially built grasping simulator and using a 15 obj
ects database. Results show that, on average, each object needed 12 success
ful experiments before an accurate grasping model was achieved. Failed expe
riments averaged to 25% of the total experiments. (C) 1998 Published by Els
evier Science Ltd. Ail rights reserved.