Lab Papers (GRASP)

Learning Tractable Word Alignment Models with Complex Constraints

João V. Graça et al. — Tue, 01 Feb 2011 10:31:40 PST

Word-level alignment of bilingual text is a critical resource for a growing variety of tasks. Probabilistic models for word alignment present a fundamental trade-off between richness of captured constraints and correlations versus efficiency and tractability of inference. In this article, we use the Posterior Regularization framework (Graça, Ganchev, and Taskar 2007) to incorporate complex constraints into probabilistic models during learning without changing the efficiency of the underlying model. We focus on the simple and tractable hidden Markov model, and present an efficient learning algorithm for incorporating approximate bijectivity and symmetry constraints. Models estimated with these constraints produce a significant boost in performance as measured by both precision and recall of manually annotated alignments for six language pairs. We also report experiments on two different tasks where word alignments are required: phrase-based machine translation and syntax transfer, and show promising improvements over standard methods.

Single Cell Manipulation using Ferromagnetic Composite Microtransporters

Mahmut Selman Sakar et al. — Thu, 14 Oct 2010 08:36:28 PDT

For biomedical applications, such as single cell manipulation, it is important to fabricate microstructures that can be powered and controlled wirelessly in fluidic environments. In this letter, we describe the construction and operation of truly micron-sized, biocompatible ferromagnetic microtransporters driven by external magnetic fields. Microtransporters were fabricated using a simple, single step fabrication method and can be produced in large numbers. We demonstrate that they can be navigated to manipulate single cells with micron-size precision without disturbing the local environment.

Temporal Logic Motion Planning for Mobile Robots

Geogios E. Fainekos et al. — Tue, 12 Oct 2010 12:49:10 PDT

In this paper, we consider the problem of robot motion planning in order to satisfy formulas expressible in temporal logics. Temporal logics naturally express traditional robot specifications such as reaching a goal or avoiding an obstacle, but also more sophisticated specifications such as sequencing, coverage, or temporal ordering of different tasks. In order to provide computational solutions to this problem, we first construct discrete abstractions of robot motion based on some environmental decomposition. We then generate discrete plans satisfying the temporal logic formula using powerful model checking tools, and finally translate the discrete plans to continuous trajectories using hybrid control. Critical to our approach is providing formal guarantees ensuring that if the discrete plan satisfies the temporal logic formula, then the continuous motion also satisfies the exact same formula.

Usability Study of a Control Framework for an Intelligent Wheelchair

Sarangi P. Parikh et al. — Tue, 12 Oct 2010 12:49:10 PDT

We describe the development and assessment of a computer controlled wheelchair called the SMARTCHAIR. A shared control framework with different levels of autonomy allows the human operator to stay in complete control of the chair at each level while ensuring her safety. The framework incorporates deliberative motion plans or controllers, reactive behaviors, and human user inputs. At every instant in time, control inputs from these three different sources are blended continuously to provide a safe trajectory to the destination, while allowing the human to maintain control and safely override the autonomous behavior. In this paper, we present usability experiments with 50 participants and demonstrate quantitatively the benefits of human-robot augmentation.

On the Time Complexity of Information Dissemination via Linear Iterative Strategies

Shreyas Sundaram et al. — Tue, 12 Oct 2010 12:49:09 PDT

Given an arbitrary network of interconnected nodes, each with an initial value, we study the number of timesteps required for some (or all) of the nodes to gather all of the initial values via a linear iterative strategy. At each time-step in this strategy, each node in the network transmits a weighted linear combination of its previous transmission and the most recent transmissions of its neighbors. We show that for almost any choice of real-valued weights in the linear iteration (i.e., for all but a set of measure zero), the number of time-steps required for any node to accumulate all of the initial values is upper-bounded by the size of the largest tree in a certain subgraph of the network; we use this fact to show that the linear iterative strategy is time-optimal for information dissemination in certain networks. In the process of deriving our results, we also obtain a characterization of the observability index for a class of linear structured systems.

Planning and Control of Mobile Robots in Image Space from Overhead Cameras

Rahul S. Rao et al. — Tue, 12 Oct 2010 12:49:09 PDT

In this work, we present a framework for the development of a planar mobile robot controller based on image plane feedback. We show that the design of such a motion controller can be accomplished in the image plane by making use of a subset of the parameters that relate the image plane to the ground plane, while still leveraging the simplifications offered by modeling the system as a differentially flat system. Our method relies on a waypoint-based trajectory generator, with all the waypoints specified in the image, as seen by an overhead observer. We present some results from simulation as well as from experiments that validate the ideas presented in this work and discuss some ideas for future work

Hybrid Controllers for Path Planning: A Temporal Logic Approach

Geogios E. Fainekos et al. — Tue, 12 Oct 2010 12:49:08 PDT

Robot motion planning algorithms have focused on low-level reachability goals taking into account robot kinematics, or on high level task planning while ignoring low-level dynamics. In this paper, we present an integrated approach to the design of closed–loop hybrid controllers that guarantee by construction that the resulting continuous robot trajectories satisfy sophisticated specifications expressed in the so–called Linear Temporal Logic. In addition, our framework ensures that the temporal logic specification is satisfied even in the presence of an adversary that may instantaneously reposition the robot within the environment a finite number of times. This is achieved by obtaining a Büchi automaton realization of the temporal logic specification, which supervises a finite family of continuous feedback controllers, ensuring consistency between the discrete plan and the continuous execution.

Learning a Manifold-Constrained Map between Image Sets: Applications to Matching and Pose Estimation

Jihun Ham et al. — Tue, 12 Oct 2010 12:49:07 PDT

This paper proposes a method for matching two sets of images given a small number of training examples by exploiting the underlying structure of the image manifolds. A nonlinear map from one manifold to another is constructed by combining linear maps locally defined on the tangent spaces of the manifolds. This construction imposes strong constraints on the choice of the maps, and makes possible good generalization of correspondences between all of the image sets. This map is flexible enough to approximate an arbitrary diffeomorphism between manifolds and can serve many purposes for applications. The underlying algorithm is a non-iterative efficient procedure whose complexity mainly depends on the number of matched training examples and the dimensionality of the manifold, and not on the number of samples nor on the dimensionality of the images. Several experiments were performed to demonstrate the potential of our method in image analysis and pose estimation. The first example demonstrates how images from a rotating camera can be mapped to the underlying pose manifold. Second, computer generated images from articulating toy figures are matched using the underlying 4 dimensional manifold to generate image-driven animations. Finally, two sets of actual lip images during speech are matched by their appearance manifold. In all these cases, our algorithm is able to obtain reasonable matches between thousands of large-dimensional images, with a minimum of computation.

Information Driven Coordinated Air-Ground Proactive Sensing

Ben Grocholsky et al. — Tue, 12 Oct 2010 12:49:07 PDT

This paper concerns the problem of actively searching for and localizing ground features by a coordinated team of air and ground robotic sensor platforms. The approach taken builds on well known Decentralized Data Fusion (DDF) methodology. In particular, it brings together established representations developed for identification and linearized estimation problems to jointly address feature detection and localization. This provides transparent and scalable integration of sensor information from air and ground platforms. As in previous studies, an Information theoretic utility measure and local control strategy drive the robots to uncertainty reducing team configurations. Complementary characteristics in terms of coverage and accuracy are revealed through analysis of the observation uncertainty for air and ground on-board cameras. Implementation results for a detection and localization example indicate the ability of this approach to scalably and effciently realize such collaborative potential.

Controlling Swarms of Robots Using Interpolated Implicit Functions

Luiz Chaimowicz et al. — Tue, 12 Oct 2010 12:49:06 PDT

We address the synthesis of controllers for large groups of robots and sensors, tackling the specific problem of controlling a swarm of robots to generate patterns specified by implicit functions of the form s(x, y) = 0. We derive decentralized controllers that allow the robots to converge to a given curve S and spread along this curve. We consider implicit functions that are weighted sums of radial basis functions created by interpolating from a set of constraint points, which give us a high degree of control over the desired 2D curves. We describe the generation of simple plans for swarms of robots using these functions and illustrate.

Control of Quantized Multi-Agent Systems with Linear Nearest Neighbor Rules: A Finite Field Approach

Shreyas Sundaram et al. — Tue, 12 Oct 2010 12:49:05 PDT

We study the problem of controlling a multi-agent system where each agent is only allowed to be in a discrete and finite set of states. Each agent is capable of updating its state based on the states of its neighbors, and there is a leader agent in the network that is allowed to update its state in arbitrary ways (within the discrete set) in order to put all agents in a desired state. We present a novel solution to this problem by viewing the discrete states of the system as elements of a finite field. Specifically, we develop a theory of structured linear systems over finite fields, and show that such systems will be controllable provided that the size of the finite field is sufficiently large, and that the graph associated with the system satisfies certain properties. We then use these results to show that a multi-agent system with a leader node is controllable via a linear nearest-neighbor update as long as there is a path from the leader to every node, and that the number of discrete states for each node is large enough.

Controlling Connectivity of Dynamic Graphs

Michael M. Zavlanos et al. — Tue, 12 Oct 2010 12:49:05 PDT

The control of mobile networks of multiple agents raises fundamental and novel problems in controlling the structure of the resulting dynamic graphs. In this paper, we consider the problem of controlling a network of agents so that the resulting motion always preserves various connectivity properties. In particular, we consider preserving k-hop connectivity, where agents are allowed to move while maintaining connections to agents that are no more than k-hops away. The connectivity constraint is translated to constrains on individual agent motion by considering the dynamics of the adjacency matrix and related constructs from algebraic graph theory. As special cases, we obtain motion constraints that can preserve the exact structure of the initial dynamic graph, or may simply preserve the usual notion connectivity while the structure of the graph changes over time. We conclude by illustrating various interesting problems that can be achieved while preserving connectivity constraints.

Control of Locomotion with Shape-Changing Wheels

Daniel Mellinger et al. — Tue, 12 Oct 2010 12:49:04 PDT

We present a novel approach to controlling the locomotion of a wheel by changing its shape, leading to applications to the synthesis and closed-loop control of gaits for modular robots. A dynamic model of a planar, continuous deformable ellipse in contact with a ground surface is derived. We present two alternative approaches to controlling this system and a method for mapping the gaits to a discrete rolling polygon. Mathematical models and dynamic simulation of the continuous approximation and the discrete n-body system, and experimental results obtained from a physical modular robot system illustrate the accuracy of the dynamic models and the validity of the approach.

Blind Sparse-nonnegative (BSN) Channel Identification for Acousitic Time-Difference-of-Arrival Estimation

Yuanqing Lin et al. — Tue, 12 Oct 2010 12:49:04 PDT

Estimating time-difference-of-arrival (TDOA) remains a challenging task when acoustic environments are reverberant and noisy. Blind channel identification approaches for TDOA estimation explicitly model multipath reflections and have been demonstrated to be effective in dealing with reverberation. Unfortunately, existing blind channel identification algorithms are sensitive to ambient noise. This papers hows how to resolve the noise sensitivity issue by exploiting prior knowledge about an acoustic room impulse response (RIR), namely, an acoustic RIR can be modeled by a sparse-nonnegative FIR filter. This paper shows how to formulate a single-input two-output blind channel identification into a least square convex optimization, and how to incorporate the sparsity and nonnegativity priors so that the resulting optimization remains convex and can be solved efficiently. The proposed blind sparse-nonnegative (BSN) channel identification approach for TDOA estimation is not only robust to reverberation, but also robust to ambient noise, as demonstrated by simulations and experiments in real acoustic environments.

Planar Ego-Motion Without Correspondences

Ameesh Makadia et al. — Tue, 12 Oct 2010 12:49:03 PDT

General structure-from-motion methods are not adept at dealing with constrained camera motions, even though such motions greatly simplify vision tasks like mobile robot localization. Typical ego-motion techniques designed for such a purpose require locating feature correspondences between images. However, there are many cases where features cannot be matched robustly. For example, images from panoramic sensors are limited by nonuniform angular sampling, which can complicate the feature matching process under wide baseline motions. In this paper we compute the planar ego-motion of a spherical sensor without correspondences. We propose a generalized Hough transform on the space of planar motions. Our transform directly processes the information contained within all the possible feature pair combinations between two images, thereby circumventing the need to isolate the best corresponding matches. We generate the Hough space in an efficient manner by studying the spectral information contained in images of the feature pairs, and by re-treating our Hough transform as a correlation of such feature pair images.

Joint covariate selection and joint subspace selection for multiple classification problems

Guillaume Obozinski et al. — Tue, 13 Oct 2009 06:37:28 PDT

We address the problem of recovering a common set of covariates that are relevant simultaneously to several classification problems. By penalizing the sum of ℓ2-norms of the blocks of coefficients associated with each covariate across different classification problems, similar sparsity patterns in all models are encouraged. To take computational advantage of the sparsity of solutions at high regularization levels, we propose a blockwise path-following scheme that approximately traces the regularization path. As the regularization coefficient decreases, the algorithm maintains and updates concurrently a growing set of covariates that are simultaneously active for all problems. We also show how to use random projections to extend this approach to the problem of joint subspace selection, where multiple predictors are found in a common low-dimensional subspace. We present theoretical results showing that this random projection approach converges to the solution yielded by trace-norm regularization. Finally, we present a variety of experimental results exploring joint covariate selection and joint subspace selection, comparing the path-following approach to competing algorithms in terms of prediction accuracy and running time.

Shape-based object recognition in videos using 3D synthetic object models

Alexander Toshev et al. — Thu, 08 Oct 2009 09:13:16 PDT

In this paper we address the problem of recognizing moving objects in videos by utilizing synthetic 3D models. We use only the silhouette space of the synthetic models making thus our approach independent of appearance. To deal with the decrease in discriminability in the absence of appearance, we align sequences of object masks from video frames to paths in silhouette space. We extract object silhouettes from video by an integration of feature tracking, motion grouping of tracks, and co-segmentation of successive frames. Subsequently, the object masks from the video are matched to 3D model silhouettes in a robust matching and alignment phase. The result is a matching score for every 3D model to the video, along with a pose alignment of the model to the video. Promising experimental results indicate that a purely shape-based matching scheme driven by synthetic 3D models can be successfully applied for object recognition in videos.

Solving Stereo Matching Problems Using Interior Point Methods

Camillo J. Taylor et al. — Wed, 07 Oct 2009 08:23:52 PDT

This paper describes an approach to reformulating the stereo matching problem as a large scale Linear Program. The approach proceeds by approximating the match cost function associated with each pixel with a piecewise linear convex function. Regularization terms related to the first and second derivative of the disparity field are also captured with piecewise linear penalty terms. The resulting large scale linear program can be tackled using interior point methods and the associated Newton Steps involve matrices that reflect the structure of the underlying pixel grid. The proposed scheme effectively exploits the structure of these matrices to solve these linear systems efficiently.

Solving Image Registration Problems Using Interior Point Methods

Camillo J. Taylor et al. — Wed, 07 Oct 2009 08:23:51 PDT

This paper describes a novel approach to recovering a parametric deformation that optimally registers one image to another. The method proceeds by constructing a global convex approximation to the match function which can be optimized using interior point methods. The paper also describes how one can exploit the structure of the resulting optimization problem to develop efficient and effective matching algorithms. Results obtained by applying the proposed scheme to a variety of images are presented.

Graph Cuts via l1 Norm Minimization

Arvind Bhusnurmath et al. — Wed, 07 Oct 2009 08:23:49 PDT

Graph cuts have become an increasingly important tool for solving a number of energy minimization problems in computer vision and other fields. In this paper, the graph cut problem is reformulated as an unconstrained l1 norm minimization that can be solved effectively using interior point methods. This reformulation exposes connections between graph cuts and other related continuous optimization problems. Eventually, the problem is reduced to solving a sequence of sparse linear systems involving the Laplacian of the underlying graph. The proposed procedure exploits the structure of these linear systems in a manner that is easily amenable to parallel implementations. Experimental results obtained by applying the procedure to graphs derived from image processing problems are provided.