Robotic Manipulation of Environmentally Constrained Objects Using Underactuated Hands (my PhD Project)
Robotics for agriculture represents the ultimate application
of one of our society's latest and most advanced innovations to its most
ancient and vital industry. Over the course of history, mechanization and
automation have increased crop output several orders of magnitude, enabling a
geometric growth in population and an increase in quality of life across the
globe. As a challenging step, manipulating objects in harvesting automation is
still under massive investigation in literature. Harvesting or the process of
gathering ripe crops can be described as breaking environmentally constrained
objects into two or more pieces at the desired locations. In this thesis, the
problem of purposefully failing (breaking) or yielding objects by a robotic
gripper is investigated. A failure task is first formulated using mechanical
failure theories. Next, a grasp quality measure is presented to characterize a
suitable grasp configuration and systematically control the failure behavior of
the object. This approach combines information about the task's failure and the
capability of the gripper for wrench insertion. The capability of the gripper
for wrench insertion can be formulated using the friction between the object
and the gripper. A new method inspired by the human pre-manipulation process is
introduced to utilize the gripper itself as the measurement tool and obtain a
friction model. The developed friction model is capable of capturing the
anisotropic behavior of materials which is the case for most fruits and
vegetables.
The limited operating space for harvesting process, the
vulnerability of agricultural products and clusters of crops demand strict
conditions for the manipulation process. This thesis presents a new sensorized
underactuated self-adaptive finger to address the stringent conditions in the
agricultural environment. This design incorporates link-driven underactuated
mechanism with an embedded load cell for contact force measurement and a
trimmer potentiometer for acquiring joint variables. The integration of these
sensors results in tactile-like sensations in the finger without compromising
the size and complexity of the proposed design. To obtain an optimum finger
design, the placement of the load cell is analyzed using Finite Element Method
(FEM). The design of the finger features a particular round shape of the distal
phalanx and specific size ratio between the phalanxes to enable both precision
and power grasps. A quantitative evaluation
of the grasp efficiency by constructing a grasp wrench space is also provided.
The effectiveness of the proposed designs and theories are
verified through real-time experiments. For conducting the massive experiments
in real-time, a software/hardware platform capable of dataset management is
crucial. In this thesis, a new comprehensive software interface for integration
of industrial robots with peripheral tools and sensors is designed and
developed. This software provides a real-time low-level access to the
manipulator controller. Furthermore, Data Acquisition boards are integrated
into the software which enables Rapid Prototyping methods. Additionally,
Hardware-in-the-loop techniques can be implemented by adding the complexity of
the plant under control to the test platform. The software is a collection of
features developed and distributed under GPL V3.0.
Control Design for Grasping Convex Objects Using Cooperative Whole Arm Manipulator (my MSc project)
This project presents a novel scale-dependent method for grasp evaluation based on so-called size ratio and the capability of the grasp to reject disturbance forces. Manipulating large objects, comparable to the robot size is considered. This is the case of a whole arm manipulation (WAM) in which the robot arm can manipulate larger objects as the whole surface is used. In this way task-oriented information is incorporated in proposed grasp evaluation criterion. Examples are presented to demonstrate the validity of the proposed approach.
Enveloping grasps are structured by wrapping the arms or fingers around the object. The main issue of the whole arm grasp (enveloping grasp) force analysis is that all contact forces are not actively controllable. It is also the case that there are always sensory errors of magnitude and position of contact forces, since object can come into contact with any part of the limbs surface. Therefore, position uncertainty results to an uncertain Jacobian matrix. We derive a new optimal adaptive Jacobian controller for regulating internal forces. Optimal contact forces are shown to be regulated with the uncertain kinematical and dynamical parameters. They are updated online by parameter update law. Experimental results validate the proposed approach.