A neural network architecture to learn the arm reach motion planning in a static cluttered environment

In this article, we present a learning model that can control a simulated anthropomorphic arm kinematics motion in order to reach and grasp a static prototypic object placed behind an obstacle of varying position and size. The network, composed of two generic neural network modules, learns to combine multi-modal arm-related information such as trajectory parameters as well as obstacle-related information such as obstacle size and location. We based our simulation to the notion of via point, which postulate that the reach motion planning is decomposed by some specifics successive position of the arm. In order to determine these particular parameters, several specifics data have been extracted from an experimental protocol and constitute the pertinent parameters which have been integrated to the model. This net of neural net determine the total path in order to reach and grasp the prototypic object avoiding the obstacle