Recursive Learning for Deformable Object Manipulation

Ayanna Howard

**
**
This research addresses the problem of robotic grasping of 3-D deformable
objects. Specifically, we seek to develop a generalized approach for
handling of 3-D deformable objects in which prior knowledge of object
attributes is not required and thus can be applied to an infinite number
of object types (e.g. asymmetric, nonhomogneeous, nonlinear object types).
Our methodology relies on the implementation of two main tasks. Our first
is to calculate deformation characteristics for a non-rigid object
represented by a physically-based model. This model is derived from
discretizing the object into a network of interconnected particles, springs,
and damping elements. Using nonlinear partial differential equations, we
model the particle motion of the deformable object in order to calculate
the deformation characteristics. For our second task, we must calculate the
the minimum force required to lift the deformable object. This lifting
force consists of a combination of the base force required to lift a rigid
object of the same weight plus an additional force term which successfully
compensates for the deformation of the object. This minimum lifting force
can be learned using a technique called iterative lifting. With this method,
the robotic system learns the required lifting force by lifting the object
with iterative measurements of force. Once the deformation characteristics
and the associated lifting force term are determined, they are used to
train a neural network for extracting the minimum force required for
subsequent deformable object manipulation tasks. Our developed algorithm
is validated with two sets of experiments. The first experimental results
are derived from the implementation of the algorithm in a simulated
environment. The second set involves a physical implementation of the
technique whose outcome is compared with the simulation results to test
the real world validity of the developed methodology. This real world
implementation consists of a dual vision system and two cooperative
manipulators, each possessing an end-effector constructed as a flat surface
palm. Based on the simulation and real-world results, we are able to show
that our physical, simulation, and theoretical lifting forces differ from
each other maximally by a 14% error level.