Showing papers by "Sebastian Thrun published in 1998"

Learning to learn

Abstract: Preface. Part I: Overview Articles. 1. Learning to Learn: Introduction and Overview S. Thrun, L. Pratt. 2. A Survey of Connectionist Network Reuse Through Transfer L. Pratt, B. Jennings. 3. Transfer in Cognition A. Robins. Part II: Prediction. 4. Theoretical Models of Learning to Learn J. Baxter. 5. Multitask Learning R. Caruana. 6. Making a Low-Dimensional Representation Suitable for Diverse Tasks N. Intrator, S. Edelman. 7. The Canonical Distortion Measure for Vector Quantization and Function Approximation J. Baxter. 8. Lifelong Learning Algorithms S. Thrun. Part III: Relatedness. 9. The Parallel Transfer of Task Knowledge Using Dynamic Learning Rates Based on a Measure of Relatedness D.L. Silver, R.E. Mercer. 10. Clustering Learning Tasks and the Selective Cross-Task Transfer of Knowledge S. Thrun, J. O'Sullivan. Part IV: Control. 11. CHILD: A First Step Towards Continual Learning M.B. Ring. 12. Reinforcement Learning with Self-Modifying Policies J. Schmidhuber, et al. 13. Creating Advice-Taking Reinforcement Learners R. Maclin, J.W. Shavlik. Contributing Authors. Index.

A Probabilistic Approach to Concurrent Mapping and Localization for Mobile Robots

Abstract: This paper addresses the problem of building large-scale geometric maps of indoor environments with mobile robots It poses the map building problem as a constrained, probabilistic maximum-likelihood estimation problem It then devises a practical algorithm for generating the most likely map from data, along with the most likely path taken by the robot Experimental results in cyclic environments of size up to 80 by 25 meter illustrate the appropriateness of the approach

The interactive museum tour-guide robot

Abstract: This paper describes the software architecture of an autonomous tour-guideltutor robot. This robot was recently deployed in the "Deutsches Museum Bonn," were it guided hundreds of visitors through the museum during a six-day deployment period. The robot's control software integrates low-level probabilistic reasoning with high-level problem solving embedded in first order logic. A collection of software innovations, described in this paper, enabled the robot to navigate at high speeds through dense crowds, while reliably avoiding collisions with obstacles--some of which could not even be perceived. Also described in this paper is a user interface tailored towards non-expert users, which was essential for the robot's success in the museum. Based on these experiences, this paper argues that time is ripe for the development of AI-based commercial service robots that assist people in everyday life.

Learning to learn: introduction and overview

Abstract: Over the past three decades or so, research on machine learning and data mining has led to a wide variety of algorithms that learn general functions from experience. As machine learning is maturing, it has begun to make the successful transition from academic research to various practical applications. Generic techniques such as decision trees and artificial neural networks, for example, are now being used in various commercial and industrial applications (see e.g., [Langley, 1992; Widrow et al., 1994]).

Lifelong learning algorithms

Abstract: Machine learning has not yet succeeded in the design of robust learning algorithms that generalize well from very small datasets. In contrast, humans often generalize correctly from only a single training example, even if the number of potentially relevant features is large. To do so, they successfully exploit knowledge acquired in previous learning tasks, to bias subsequent learning.

Learning to classify text from labeled and unlabeled documents

Abstract: In many important text classification problems, acquiring class labels for training documents is costly, while gathering large quantities of unlabeled data is cheap. This paper shows that the accuracy of text classifiers trained with a small number of labeled documents can be improved by augmenting this small training set with a large pool of unlabeled documents. We present a theoretical argument showing that, under common assumptions, unlabeled data contain information about the target function. We then introduce an algorithm for learning from labeled and unlabeled text based on the combination of Expectation-Maximization with a naive Bayes classifier. The algorithm first trains a classifier using the available labeled documents, and probabilistically labels the unlabeled documents; it then trains a new classifier using the labels for all the documents, and iterates to convergence. Experimental results, obtained using text from three different realworld tasks, show that the use of unlabeled data reduces classification error by up to 33%.

Map learning and high-speed navigation in RHINO

Abstract: This chapter surveys basic methods for learning maps and high speed autonomous navigation for indoor mobile robots. The methods have been developed in our lab over the past few years, and most of them have been tested thoroughly in various indoor environments. The chapter is targeted towards researchers and engineers who attempt to build reliable mobile robot navigation software.

Bayesian Landmark Learning for Mobile Robot Localization

Abstract: To operate successfully in indoor environments, mobile robots must be able to localize themselves. Most current localization algorithms lack flexibility, autonomy, and often optimality, since they rely on a human to determine what aspects of the sensor data to use in localization (e.g., what landmarks to use). This paper describes a learning algorithm, called BaLL, that enables mobile robots to learn what features/landmarks are best suited for localization, and also to train artificial neural networks for extracting them from the sensor data. A rigorous Bayesian analysis of probabilistic localization is presented, which produces a rational argument for evaluating features, for selecting them optimally, and for training the networks that approximate the optimal solution. In a systematic experimental study, BaLL outperforms two other recent approaches to mobile robot localization.

Integrating topological and metroc maps for mobile robot navigation: a statistical approach

Abstract: The problem of concurrent mapping and localization has received considerable attention in the mobile robotics community. Existing approaches can largely be grouped into two distinct paradigms: topological and metric. This paper proposes a method that integrates both. It poses the mapping problem as a statistical maximum likelihood problem, and devises an efficient algorithm for search in likelihood space. It presents an novel mapping algorithm that integrates two phases: a topological and a metric mapping phase. The topological mapping phase solves a global position alignment problem between potentially indistinguishable, significant places. The subsequent metric mapping phase produces a fine-grained metric map of the environment in floating-point resolution. The approach is demonstrated empirically to scale up to large, cyclic, and highly ambiguous environments.

Finding landmarks for mobile robot navigation

Abstract: Localization addresses the problem of determining the position of a mobile robot from sensor data. This paper presents an algorithm, called BaLL, which enables a mobile robot to learn a set of landmarks used in localization and to learn how to recognize them using artificial neural networks. BaLL is based on a statistical localization approach. It is applicable to a large variety of sensors and environments. Experiments with a mobile robot equipped with sonar sensors and a camera illustrate that BaLL identifies highly useful landmarks.

Position estimation for mobile robots in dynamic environments

Abstract: For mobile robots to be successful, they have to navigate safely in populated and dynamic environments. While recent research has led to a variety of localization methods that can track robots well in static environments, we still lack methods that can robustly localize mobile robots in dynamic environments, in which people block the robot's sensors for extensive periods of time or the position of furniture may change. This paper proposes extensions to Markov localization algorithms enabling them to localize mobile robots even in densely populated environments. Two different filters for determining the "believability" of sensor readings are employed. These filters are designed to detect sensor readings that are corrupted by humans or unexpected changes in the environment. The technique was recently implemented and applied as part of an installation, in which a mobile robot gave interactive tours to visitors of the "Deutsches Museum Bonn." Extensive empirical tests involving datasets recorded during peak traffic hours in the museum demonstrate that this approach is able to accurately estimate the robot's position in more than 98% of the cases even in such highly dynamic environments.

A hybrid collision avoidance method for mobile robots

Abstract: Proposes a hybrid approach to the problem of collision avoidance for indoor mobile robots. The /spl mu/DWA (model-based dynamic window approach) integrates sensor data from various sensors with information extracted from a map of the environment, to generate collision-free motion. A novel integration rule ensures that with high likelihood, the robot avoids collisions with obstacles not detectable with its sensors, even if it is uncertain about its position. The approach was implemented and tested extensively as part of an installation, in which a mobile robot gave interactive tours to visitors of the "Deutsches Museum Bonn." Here our approach was essential for the success of the entire mission, because a large number of ill-shaped obstacles prohibited the use of purely sensor-based methods for collision avoidance.

Exploration in active learning

Abstract: In the passive learning paradigm, a learner learns purely through observing its environment. The environment is assumed to generate a stream of training data according to some unknown probability distribution. Passive learning techniques di er in the type of results they seek to produce, as well as in the way they generalize from observations. Common learning tasks are the clustering, classi cation, or prediction of future data.

Clustering learning tasks and the selective cross-task transfer of knowledge

Abstract: Recently, there has been an increased interest in machine learning methods that transfer knowledge across multiple learning tasks and “learn to learn.” Such methods have repeatedly been found to outperform conventional, single-task learning algorithms when the learning tasks are appropriately related. To increase robustness of such approaches, methods are desirable that can reason about the relatedness of individual learning tasks, in order to avoid the danger arising from tasks that are unrelated and thus potentially misleading.

Probabilistic mapping of an environment by a mobile robot

Abstract: This paper addresses the problem of building large-scale maps of indoor environments with mobile robots. It proposes a statistical approach that describes the map building problem as a constrained maximum-likelihood estimation problem, for which it devises a practical algorithm. Experimental results in large, cyclic environments illustrate the appropriateness of the approach.

Using EM to Classify Text from Labeled and Unlabeled Documents

Abstract: : This paper shows that the accuracy of learned text classifiers can be improved by augmenting a small number of labeled training documents with a large pool of unlabeled documents. This is significant because in many important text classification problems obtaining classification labels is expensive, while large quantities of unlabeled documents are readily available. We present a theoretical argument showing that, under common assumptions, unlabeled data contain information about the target function. We then introduce an algorithm for learning from labeled and unlabeled text, based on the combination of Expectation-Maximization with a naive Bayes classifier. The algorithm first trains a classifier using the available labeled documents, and probabilistically labels the unlabeled documents. It then trains a new classifier using the labels for all the documents, and iterates. Experimental results, obtained using text from three different real-world tasks, show that the use of unlabeled data reduces classification error by up to 30%.

Jacobian images of super-resolved texture maps for model-based motion estimation and tracking

Abstract: We present a Kalman filter based approach to perform model-based motion estimation and tracking. Unlike previous approaches, the tracking process is not formulated as an SSD minimization problem, but is developed by using texture mapping as the measurement model in an extended Kalman filter. During tracking, a super-resolved estimate of the texture present on the object or in the scene is obtained. A key result is the notion of Jacobian images, which can be viewed as a generalization of traditional gradient images, and represent the crucial computation in the tracking process. The approach is illustrated with three sample applications: full 3D tracking of planar surface patches, a projective surface tracker for uncalibrated camera scenarios, and a fast, Kalman filtered version of mosaicking with detection of independently moving objects.

Towards exact localization without explicit localization with the generalized Voronoi graph

Abstract: Sensor based exploration is a task which enables a robot to explore and map an unknown environment, using sensor information. The map used in this paper is the generalized Voronoi graph (GVG). The robot explores an unknown environment using an already developed incremental construction procedure to generate the GVG using sensor information. This paper presents some initial results which uses the GVG for robot localization, while mitigating the need to update encoder values. Experimental results verify the described work.

Super-resolved texture tracking of planar surface patches

Abstract: We present an approach to tracking planar surface patches over time. In addition to tracking a patch with full six degrees of freedom, the algorithm also produces a super-resolved estimate of the texture present on the patch. This texture estimate is kept as an explicit model texture image which is refined over time. We then use it to infer the 3D motion of the patch from the image sequence. The main idea behind the approach is to use a technique from computer graphics, known as texture mapping, as the measurement model in an extended Kalman filter. We also calculate the partial derivative of this image formation process with respect to the 3D pose of the patch, which functions as the measurement Jacobian. The super-resolved estimate of the texture is obtained using the standard extended Kalman filter measurement update, with one essential approximation that makes this computationally feasible. The resulting equations are remarkably simple, yet lead to estimates that are properly super-resolved. In addition to developing the theory behind the approach, we also demonstrate both the tracking and the super-resolution aspect of the algorithm on real image sequences.

Template-based recognition of pose and motion gestures on a mobile robot

Abstract: For mobile robots to assist people in everyday life, they must be easy to instruct. This paper describes a gesture-based interface for human robot interaction, which enables people to instruct robots through easy-to-perform arm gestures. Such gestures might be static pose gestures, which involve only a specific configuration of the person's arm, or they might be dynamic motion gestures (such as waving). Gestures are recognized in real-time at approximate frame rate, using a hybrid approach that integrates neural networks and template matching. A fast, color-based tracking algorithm enables the robot to track and follow a person reliably through office environments with drastically changing lighting conditions. Results are reported in the context of an interactive clean-up task, where a person guides the robot to specific locations that need to be cleaned, and the robot picks up trash which it then delivers to the nearest trash-bin.

Learning to Locate an Object in 3D Space from a Sequence of Camera Images

Abstract: This paper addresses the problem of determining an object’s 3D location from a sequence of camera images recorded by a mobile robot. The approach presented here allows people to “train” robots to recognize specific objects, by presenting it examples of the object to be recognized. A decision tree method is used to learn significant features of the target object from individual camera images. Individual estimates are integrated over time using Bayes rule, into a probabilistic 3D model of the robot’s environment. Experimental results illustrate that the method enables a mobile robot to robustly estimate the 3D location of objects from multiple camera images.

A Framework for Programming Embedded Systems: Initial Design and Results

Abstract: : This paper describes CES, a prototype of a new programming language for robots and other embedded systems, equipped with sensors and actuators. CES contains two new ideas, currently not found in other programming languages: support of computing with uncertain information, and support of adaptation and teaching as a means of programming. These innovations facilitate the rapid development of software for embedded systems, as demonstrated by two mobile robot applications.

When robots meet people

Abstract: An ongoing revolution is taking place in mobile robotics. Robots that guide blind or mentally handicapped people, clean large office buildings and department stores, or assist people in recreational activities are clearly within reach. For many of those target domains, prototypes are readily available. The author discusses a case study of a mobile tour guide robot in a branch of the Deutsches Museum. The robot's task was to guide visitors through the museum, explaining to them a subset of the museum's exhibits. The project's major challenges are grouped into two categories: navigation and human-robot interaction.

Monte Carlo Hidden Markov Models

Abstract: : We present a learning algorithm for hidden Markov models with continuous state and observation spaces. All necessary probability density functions are approximated using samples, along with density trees generated from such samples. A Monte Carlo version of Baum-Welch (EM) is employed to learn models from data, just as in regular HMM learning. Regularization during learning is obtained using an exponential shrinking technique. The shrinkage factor, which determines the effective capacity of the learning algorithm, is annealed down over multiple iterations of Baum-Welch, and early stopping is applied to select the right model. We prove that under mild assumptions, Monte Carlo Hidden Markov Models converge to a local maximum in likelihood space, just like conventional HMMs. In addition, we provide empirical results obtained in a gesture recognition domain, which illustrate the appropriateness of the approach in practice.

A neural-network based approach for recognition of pose and motion gestures on a mobile robot

Abstract: Since a variety of changes in both robotic hardware and software suggests that service robots will soon become possible, to find "natural" ways of communication between human and robots is of fundamental importance for the robotic field. The paper describes a gesture-based interface for human-robot interaction, which enables people to instruct robots through easy-to-perform arm gestures. Such gestures might be static pose gestures, which involve only a specific configuration of the person's arm, or they might be dynamic motion gestures, that is, they involve motion (such as waving). Gestures are recognized in real-time at approximate frame rate, using neural networks. A fast, color-based tracking algorithm enables the robot to track and follow a person reliably through office environments with drastically changing lighting conditions. Results are reported in the context of an interactive clean-up task, where a person guides the robot to specific locations that need to be cleaned, and the robot picks up trash which it then delivers to the nearest trash-bin.

Markov Localization for Reliable Robot Navigation and People Detection

Abstract: Localization is one of the fundamental problems in mobile robotics. Without knowledge about their position mobile robots cannot efficiently carry out their tasks. In this paper we present Markov localization as a technique for estimating the position of a mobile robot. The key idea of this technique is to maintain a probability density over the whole state space of the robot within its environment. This way our technique is able to globally localize the robot from scratch and even to recover from localization failures, a property which is essential for truly autonomous robots. The probabilistic framework makes this approach robust against approximate models of the environment as well as noisy sensors. Based on a fine-grained, metric discretization of the state space, Markov localization is able to incorporate raw sensor readings and does not require predefined landmarks. It also includes a filtering technique which allows to reliably estimate the position of a mobile robot even in densely populated environments. We furthermore describe, how the explicit representation of the density can be exploited in a reactive collision avoidance system to increase the robustness and reliability of the robot even in situations in which it is uncertain about its position. The method described here has been implemented and tested in several real-world applications of mobile robots including the deployments of two mobile robots as interactive museum tour-guides.