We present an approach to transductive learning that employs semi-supervised clustering of all available data (both labeled and unlabeled) to produce a data-dependent SVM kernel. In the general case where the domain includes irrelevant or redundant attributes, we constrain the clustering to occur on the manifold prescribed by the data (both labeled and unlabeled). Empirical results show that the approach performs comparably to more traditional kernels while providing significant reduction in the number of support vectors used. Further, the kernel construction technique provides some of the benefits that would normally be provided by dimensionality reduction preprocessing step.

Modern approaches to simultaneous localization and mapping (SLAM) formulate the inference problem as a high-dimensional but sparse nonconvex M-estimation, and then apply general first- or second-order smooth optimization methods to recover a local minimizer of the objective function. The performance of any such approach depends crucially upon initializing the optimization algorithm near a good solution for the inference problem, a condition that is often difficult or impossible to guarantee in practice. To address this limitation, in this paper we present a formulation of the SLAM M-estimation with the property that, by expanding the feasible set of the estimation program, we obtain a convex relaxation whose solution approximates the globally optimal solution of the SLAM inference problem and can be recovered using a smooth optimization method initialized at any feasible point. Our formulation thus provides a means to obtain a high-quality solution to the SLAM problem without requiring high-quality initialization.

Online video presents a great opportunity for up-and-coming singers and artists to be visible to a worldwide audience. However, the sheer quantity of video makes it difficult to discover promising musicians. We present a novel algorithm to automatically identify talented musicians using machine learning and acoustic analysis on a large set of "home singing" videos. We describe how candidate musician videos are identified and ranked by singing quality. To this end, we present new audio features specifically designed to directly capture singing quality. We evaluate these vis-a-vis a large set of generic audio features and demonstrate that the proposed features have good predictive performance. We also show that this algorithm performs well when videos are normalized for production quality.

We describe our efforts to create infrastructure to enable web interfaces for robotics. Such interfaces will enable researchers and users to remotely access robots through the internet as well as expand the types of robotic applications available to users with web-enabled devices. This paper centers on rosjs, a lightweight Javascript binding for ROS, Willow Garage's robot middleware framework, rosjs exposes many of the capabilities of ROS, allowing application developers to write controllers that are executed through a web browser. We discuss how rosjs extends ROS and briefly overview some of the features it provides, rosjs has been instrumental in the creation of remote laboratories featuring the iRobot Create and the PR2. These facilities will be available to the community as experimental resources. We describe the overall goals of this project as well as provide a brief description of how rosjs was used to help create web interfaces for these facilities.

Reliability and availability are major concerns for autonomous systems. A personal robot has to solve complex tasks, such as loading a dishwasher or folding laundry, which are very difficult to automate robustly. In order for a robot to perform better in those applications, it needs to be capable of accepting help from a human operator. Shared autonomy is a system model based on human-robot dialogue. This work aims at bridging the gap between full human control and full autonomy for tasks in the domain of personal robotics. One of the hardest problems for personal robotic systems is perception: perceiving and inferring about objects in the robot's environment. We present a system capable of solving the perceptual inference in combination with a human, such that a human operator functions as a resource for the robot and helps to compensate for limitations of autonomy. In this paper, we show how a human-robot team can work together effectively to solve complex perception tasks. We present a system that asks a human operator to identify objects it doesn't recognize or find. In various experiments with the PR2 robot we show that this shared autonomy system performs more robustly than the robot system alone and that it is capable of tasks which are difficult to accomplish by an autonomous agent.

This paper describes a system for automatic mapping and generation of textured 3D models of indoor environments without user interaction. Our data acquisition system is based on a Segway RMP platform which allows us to automatically acquire large amounts of textured 3D scans in a short amount of time. The first data processing step is registration and mapping. We propose a probabilistic, non-rigid registration method that incorporates statistical sensor models and surface prior distributions to optimize alignment and the reconstructed surface at the same time. Second, in order to fuse multiple scans and to reconstruct a consistent 3D surface representation, we incorporate a volumetric surface reconstruction method based on a oriented point. For the final step of texture reconstruction, we present a novel method to automatically generate blended textures from multiple images and multiple scans which are mapped onto the 3D model for photo-realistic visualization. We conclude our report with results from a large-scale, real-world experiment. The most significant contribution of this research is a functional system that covers all steps required to automatically reconstruct textured 3D models of large indoor environments.

We consider the task of learning to accurately follow a trajectory in a vehicle such as a car or helicopter. A number of dynamic programming algorithms such as Differential Dynamic Programming (DDP) and Policy Search by Dynamic Programming (PSDP), can efficiently compute non-stationary policies for these tasks --- such policies in general are well-suited to trajectory following since they can easily generate different control actions at different times in order to follow the trajectory. However, a weakness of these algorithms is that their policies are time-indexed, in that they apply different policies depending on the current time. This is problematic since 1) the current time may not correspond well to where we are along the trajectory and 2) the uncertainty over states can prevent these algorithms from finding any good policies at all. In this paper we propose a method for space-indexed dynamic programming that overcomes both these difficulties. We begin by showing how a dynamical system can be rewritten in terms of a spatial index variable (i.e., how far along the trajectory we are) rather than as a function of time. We then use these space-indexed dynamical systems to derive space-indexed version of the DDP and PSDP algorithms. Finally, we show that these algorithms perform well on a variety of control tasks, both in simulation and on real systems.