This thesis presents methods to improve the manipulation capabilities of robots. People and animals can effectively handle objects of many shapes, sizes, weights, and materials using a variety of manipulation primitives such as grasping, pushing, sliding, tipping, rolling, and throwing. In contrast, most robots manipulate objects by pick-and-place. Restricting robots to only grasp objects artificially limits the set of tasks that they can accomplish, and leveraging a larger set of manipulation primitives is crucial for robots to reach their full potential in applications such as flexible manufacturing, agricultural automation, service industries, and disaster response. We first outline a high-level framework for planning and control for dynamic, graspless, and hybrid manipulation tasks. "Dynamic" means that the momentum of the objects cannot be ignored. "Graspless" means that the objects are not grasped in a traditional sense, but rather manipulated with unilateral friction forces at the contacts. "Hybrid" means that multiple manipulation primitives are used that each have their own constraints and dynamic equations. The main contributions of this work are the framework that outlines specific subproblems to solve dynamic, graspless, and hybrid manipulation tasks as well as an experimental implementation for an example task. In the remaining chapters we focus specifically on modeling rolling contacts between two smooth bodies, designing motion planners and feedback controllers for rolling manipulation tasks, and testing them in simulation and experimentally. First-order kinematics addresses the rolling problem where the relative contact velocities are given. The second-order kinematics is a generalization of the first-order model where the relative accelerations at the contact are specified. The evolution of dynamic rolling systems is governed by forces and torques at the contact. We address both first- and second-order kinematics and dynamic rolling in our work. The main contributions of the first-order rolling work are a robust motion planner that can handle general smooth geometries, a method to test the controllability of linearized rolling trajectories, and a method to stabilize trajectories from initial state perturbations. The main contributions of the second-order rolling work are corrections to previous work that derived the second-order kinematics equations. The main contributions of the dynamic rolling work are that it is the first work we know of to formulate the rolling dynamics of a rigid body rolling on a six degree-of-freedom motion-controlled manipulator for general manipulator and object shapes, it provides a compact form that outputs the dynamics and contact forces that can be used for trajectory optimization, and the coordinate-based representation allows the dynamics to be linearized to generate feedback controllers that stabilize planned trajectories. We apply the method to model, plan, and stabilize dynamic, graspless, and hybrid rolling experiments that include the first known implementation of the rolling pendulum swing up.

Creator

• Woodruff, James Zachary

DOI

• https://doi.org/10.21985/n2-scrk-h789

Subject

• Robotics

Language

• en

Alternate Identifier

• etdadmin_upload_781100
• http://dissertations.umi.com/northwestern:15422

Keyword

• Manipulation
• Controls
• Hybrid
• Robotic
• Motion Planning

Date created

• 2020-01-01

Resource type

• Dissertation

Rights statement

• In Copyright