Information Theory for Arm Visuo-Motor Coordination

166 p.Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 1993.The present work addresses the information processing of visuo-motor coordination. The aim of this research is to develop an adaptive model for coordinating mechanical action of an arm according to visual information. As a result, a large neural map has been generated which has been used to guide a pneumatically driven robot arm through a vision system consisting of image boards and two stereo cameras. This engineering approach has been employed to test the algorithm which might share close features with the way that biological beings solve the same sensory-motor task.Basically, arm postures are represented through their projections onto a set of image planes. Based on the link orientations and lengths as visual primitives extracted from these images, a topological state-space is characterized. Arm kinematics is defined as transformations of topological hypersurfaces, the intersections of which represent all possible postures which any redundant arm possesses in reaching an arbitrary target position. The self-organizing feature map has learned how the topological hypersurfaces transform in the state-space during arbitrary movements of the arm. The analyses of these transformations helped in idealizing a connectionistic model for kinematics.A model for the collision-free motion of a redundant arm manipulator moving in a workspace with obstacles is presented. A mapping of the arm surface onto a set of lattices, in which visual, motor information, and surface location are encoded, is adaptively developed through a learning procedure fed by trial movements. The map, which carries topographical features of the arm surface, is then used to guide the arm avoiding collisions with obstacles in unpredictable positions.The main achievements of this research are the topographical neural model for obstacle-avoidance and the connectionist model for kinematics for redundant arms. Both models have been developed based on the analyses of a dynamic geometry, induced by the arm movements, embedded in a denoted visuo-motor space. The topographical neural model presents similar features with the motor cortex which might provide some insight for understanding biological visuo-motor control.U of I OnlyRestricted to the U of I community idenfinitely during batch ingest of legacy ETD