Publications

Abstract not provided.

Abstract not provided.

Abstract not provided.

Modeling geospatial information with semantic graphs enables search for sites of interest based on relationships between features, without requiring strong a priori models of feature shape or other intrinsic properties. Geospatial semantic graphs can be constructed from raw sensor data with suitable preprocessing to obtain a discretized representation. This report describes initial work toward extending geospatial semantic graphs to include temporal information, and initial results applying semantic graph techniques to SAR image data. We describe an efficient graph structure that includes geospatial and temporal information, which is designed to support simultaneous spatial and temporal search queries. We also report a preliminary implementation of feature recognition, semantic graph modeling, and graph search based on input SAR data. The report concludes with lessons learned and suggestions for future improvements.

Abstract not provided.

This paper presents an implemented algorithm that automatically designs fixtures and assembly pallets to hold three-dimensional parts. The designed fixtures rigidly constrain and locate the part, obey task constraints, are robust to part shape variations, are easy to load, and are economical to produce. The algorithm is guaranteed to find the global optimum solution that satisfies these and other pragmatic conditions. We present the results of the algorithm applied to several practical manufacturing problems. For these complex problems the algorithm typically returns initial high-quality fixture designs in less than two minutes, and identifies th global optimum design in just over an hour.

Costs associated with designing and fabricating fixtures may be a significant portion of the total costs associated with a manufacturing task. The software tool, HoldFast, designs optimal fixtures that hold a single workpiece, are easily fabricated, provide rigid constraint and deterministic location of the workpiece, are robust to workpiece shape variations, obey all associated task constraints, and are easy to load and unload. We illustrate the capabilities of HoldFast by designing fixtures for several examples. Fixtures are designed and built for finish-machining and drilling of a cast part for prototype fabrication and mass-production fabrication. A pallet fixture is designed for vertical assembly of a personal cassette player. Another pallet fixture is designed and built that will hold either the personal cassette player or a glue gun during assembly.

Successful robot systems must employ actions that are robust in the face of task uncertainty. Toward this end, Lozano-Perez, Mason, and Taylor developed a model of manipulation tasks that explicitly considers task uncertainty. In this paper we study the utility of this model applied to real-world tasks. We report the results of two experiments that highlight the strengths and weaknesses of the LMT approach. The first experiment showed that the LMT formalism can successfully plan solutions for a complex real-world task. The second experiment showed a task that the formalism is fundamentally incapable of solving.

A key feature distinguishing robotics from traditional computer science is its connection to the physical world. Robot planning software may use elegant algorithms supported by ironclad analytic proofs, but ultimately nature will decide whether the software output is correct in the sense of accomplishing the task goal. Thus a chief goal of robotics research is to understand and capture this nature in a way that allows algorithmic analysis to produce robust physical results. This is made particularly difficult by the presence of uncertainty, which arises from the inevitable discrepancy between the real task and its idealized computer model. This paper reviews fundamental sets of states, forces, and actions that exist for a broad class of robot manipulation tasks, and ties these sets to past and future approaches to developing robust manipulation planning and execution systems.

This paper addresses the problem of manipulation planning in the presence of uncertainty. We begin by reviewing the worst-case planning techniques introduced in and show that these methods are hampered by an information gap inherent to worst-case analysis techniques. As the task uncertainty increases, these methods fail to produce useful information even though a high-quality plan may exist. To fill this gap, we present the probabilistic backprojection, which describes the likelihood that a given action will achieve the task goal from a given initial state. We provide a constructive definition of the probabilistic backprojection and related probabilistic models of manipulation task mechanics, and show how these models unify and enhance several past results in manipulation planning. These models capture the fundamental nature of the task behavior, but appear to be very complex. Methods for computing these models are sketched, but efficient computational methods remain unknown.

This paper presents two algorithms that construct a set of initial (x, y, {theta}) configurations from which a given action will reliably accomplish a planar manipulation task. The first algorithm applies energy arguments to construct a conservative set of successful initial configurations, while the second algorithm performs numerical integration to construct a set that is much less conservative. The algorithms may be applied to a variety of tasks, including pushing, placing-by-dropping, and force-controlled assembly tasks. Both algorithms consider the task geometry and mechanics, and allow uncertainty in every task parameter except for the object shapes. Experimental results are presented which demonstrate the validity of the algorithms' output for two example manipulation tasks. 16 refs., 8 figs.