Abstract
D.Phil. (Mechanical Engineering)
Grasping is the ability to seize or hold an object. Robotic grasping presents
many unsolved challenges. These challenges are solved, through evolution
for primate grasping. This study begins with the premise that there are insights
from neuroscience that can be leveraged to inform robotic grasping.
The last few years of research have provided deeper insight into the cortical
grasping areas of the macaque parietal areas including the Caudal and Anterior
Intraparietal (CIP and AIP) and pre-motor F5ab area. We now know that
these areas are key for planning and execution of motor actions for grasping.
The CIP is located on the dorsal pathway, it receives depth information from
the visual cortex and calculates orientation and curvature information of a
presented object to AIP. The AIP area (Brodmann area 7) receives input from
the visual cortex and is involved in visually guided grasping and provides
specific shape information for hand grasping movements. Area F5ab manages
control of the end-effectors pre-shape position by transforming shape
and size calculations provided by the AIP area. We use this understanding
to delineate stimuli necessary for grasping. The CIP, AIP and F5ab were set
as the grasping network for this investigation.
With the neuroscience of grasping in mind. The tools of deep learning are
discovered here in an attempt to emulate the primate grasping network. A
model for areas CIP and AIP is developed directly from electrophysiology
data, and the concepts of the grasping network are used to build a deep
learning network...