the author and should not be interpreted as representing the o cial policies, either expressed or A good model of the mechanics of a task is a resource for a robot, just as actuators and sensors are resources. The e ective use of frictional, gravitational, and dynamic forces can substitute for extra actuators; the expectation derived from a good model can minimize sensing requirements. Despite this, most robot systems attempt to dominate or nullify task mechanics, rather than exploit them. There has been little e ort to understand the manipulation capabilities of even the simplest robots under more complete mechanics models. This thesis addresses that knowledge de cit by studying graspless or nonprehensile manipulation. Nonprehensile manipulation exploits task mechanics to achieve a goal state without grasping, allowing simple mechanisms to accomplish complex tasks. With nonprehensile manipulation, a robot can manipulate objects too large or heavy to be grasped and lifted, and a low-degree-of-freedom robot can control more degrees-of-freedom of an object by allowing relative motion between the object and the manipulator. Two key problems are determining controllability of and motion planning for

We consider the path-planning problem for a robot pushing an object in an environment containing ob-stacles. This new variant of the classical robot path-planning problem has several interesting geometric aspects, which we explore. We focus on the setting where the robot makes a point contact with the ob-ject which is assumed to be a unit disk, while the obstacles are assumed to be polygonal. 1

Abstract|We describe an unsupervised on-line method for learning of manipulative actions that allows a robot to push an object connected to it with a rotational point contact to a desired point in image-space. By observing the results of its actions on the object's orientation in imagespace, the system forms a predictive forward empirical model. This acquired model is used on-line for manipulation planning and control as it improves. Rather than explicitly inverting the forward model to achieve trajectory control, a stochastic action selection technique [Moore, 1990] is used to select the most informative and promising actions, thereby integrating active perception and learning by combining on-line improvement, task-directed exploration, and model exploitation. Simulation and experimental results of the approach are presented. I.

Abstract — We develop a systematic approach to incorporating uncertainty into planning manipulation tasks with frictional contacts. We consider the canonical problem of assembling a peg into a hole at the meso scale using probes with minimal actuation but with visual feedback from an optical microscope. We consider three sources of uncertainty. Because of errors in sensing position and orientation of the parts to be assembled, we must consider uncertainty in the sensed configuration of the system. Second, there is uncertainty because of errors in actuation. Third, there are geometric and physical parameters characterizing the environment that are unknown. We discuss the synthesis of robust planning primitives using a single degreeof-freedom probe and the automated generation of plans for meso-scale manipulation. We show simulation and experimental results in support of our work. I.

This paper introduces a framework for the task of observing and controlling the movement of large objects by the use of pushing manipulation in the presence of obstacles. Accomplishing this task requires that sub-tasks behaviors be executed simultaneously. The control of the sub-tasks, based on Discrete Event Dynamic System Theory (DEDS), is accomplished by introducing a task supervisor. Its role is that of controlling and arbitrating the different behaviors to guarantee correct execution of the task. The issuing of various control patterns, or commands, to the agent pushing the object, is a consequence of feed-back obtained from the environment through one or more observers. The individual behaviors are driven by qualitative measurements from the visual space rather than explicit full three dimensional reconstruction of the space around the pushing agents. Hence real-time dynamic control can be performed by the individual behaviors and constraints are imposed and modulated by the inte...

We describe an unsupervised on-line method for learning of manipulative actions that allows a robot to push an object connected to it with a rotational point contact to a desired point in image-space. By observing the results of its actions on the object's orientation in imagespace, the system forms a predictive forward empirical model. This acquired model is used on-line for manipulation planning and control as it improves. Rather than explicitly inverting the forward model to achieve trajectory control, a stochastic action selection technique [Moore, 1990] is used to select the most informative and promising actions, thereby integrating active perception and learning by combining on-line improvement, task-directed exploration, and model exploitation. Simulation and experimental results of the approach are presented. I. INTRODUCTION Active perception can broadly be defined as the process of information gathering, organization and interpretation by the active and purposive control of...

This paper describes a method for robotic manipulation that uses direct image-space calculation of optical flow information for continuous real-time control of manipulative actions. State variables derived from optical flow measurements are described. The resulting approach is advantageous since it robustifies the system to changes in optical parameters and also simplifies the implementation needed to succeed in the task execution. Two reference tasks and their corresponding experiments are described: the insertion of a pen into a "cap" (the capping experiment) and the rotational point-contact pushing of an object of unknown shape, mass and friction to a specified goal point in the image-space. I. INTRODUCTION The visual system of an agent, either natural or artificial, has to cope with motion in at least two ways: it should be able to detect, measure and interpret the motion of external objects, and it must be able to use dynamic visual information to control, plan and coordinate its...

Abstract — Object delivery by pushing objects with mobile robots on a flat surface has been successfully demonstrated. However, existing methods can push objects that have a circular or rectangular shape. In this paper, we introduce a learningbased approach for pushing objects of any irregular shape to user-specified goal locations. We first automatically collect a set of data on how an irregular-shaped object moves given the robot’s relative position and pushing direction. We collect this data with a randomized approach, and we demonstrate that this approach can successfully collect useful data. Object delivery is achieved by using the collected data with a nonparametric regression method. We demonstrate our approach with a number of irregular-shaped objects. I.