Refine
Has Fulltext
- no (17)
Document Type
- Conference Proceeding (13)
- Article (2)
- Part of a Book (2)
Language
- English (17)
Is part of the Bibliography
- no (17)
This paper presents a framework for performing real-time recursive estimation of landmarks’ visual appearance. Imaging data in its original high dimensional space is probabilistically mapped to a compressed low dimensional space through the definition of likelihood functions. The likelihoods are subsequently fused with prior information using a Bayesian update. This process produces a probabilistic estimate of the low dimensional representation of the landmark visual appearance. The overall filtering provides information complementary to the conventional position estimates which is used to enhance data association. In addition to robotics observations, the filter integrates human observations in the appearance estimates. The appearance tracks as computed by the filter allow landmark classification. The set of labels involved in the classification task is thought of as an observation space where human observations are made by selecting a label. The low dimensional appearance estimates returned by the filter allow for low cost communication in low bandwidth sensor networks. Deployment of the filter in such a network is demonstrated in an outdoor mapping application involving a human operator, a ground and an air vehicle.
This paper considers the modes of interaction between one or several human operators and an active sensor network-a fully decentralized network of sensors some or all of which have actuators and are in that sense active. The primary goal of this study is to investigate the conditions under which the human involvement will not jeopardize scalability of the overall system. Two aspects of human-robot interaction are considered: the ways in which the global view of the system may be conveyed to the operators, and how the operators may influence the behavior of the system during the course of its operation. The results of analysis favor peer-to-peer environment-centric interactions between the operators and the network whereby the humans act as extended sensors and communication nodes of the network itself. Experiments on an indoor active sensor network are described.
In this paper, we present the application of a non-linear dimensionality reduction technique for the learning and probabilistic classification of hyperspectral image. Hyperspectral image spectroscopy is an emerging technique for geological investigations from airborne or orbital sensors. It gives much greater information content per pixel on the image than a normal colour image. This should greatly help with the autonomous identification of natural and manmade objects in unfamiliar terrains for robotic vehicles. However, the large information content of such data makes interpretation of hyperspectral images time-consuming and userintensive. We propose the use of Isomap, a non-linear manifold learning technique combined with Expectation Maximisation in graphical probabilistic models for learning and classification. Isomap is used to find the underlying manifold of the training data. This low dimensional representation of the hyperspectral data facilitates the learning of a Gaussian Mixture Model representation, whose joint probability distributions can be calculated offline. The learnt model is then applied to the hyperspectral image at runtime and data classification can be performed.
This paper presents the design of a probabilistic model of human perception as an integral part of a decentralized data fusion system. The system consists of a team of human operators and robotic platforms, together forming a heterogeneous sensor network. Human operators are regarded as information sources submitting raw observations. The observations are converted into a probabilistic representation suitable for fusion with the system's belief. The conversion is performed by a human sensor model (HSM). The initial HSM is built offline based on an average of multiple human subjects conducting a calibration experiment. Since individual human operators may vary in their performance, an online adaptation of the HSM is required. The network estimate is used for adaptation because the true feature state is unknown at runtime. Results of an outdoor calibration experiment using range and bearing observations are presented. Simulations show the feasibility of efficient online adaptation.
This paper presents the development and demonstration of non-Gaussian, decentralised state estimation using an outdoor sensor network consisting of an autonomous air vehicle, a manual ground vehicle, and two human operators. The location and appearance of landmarks were estimated using bearing only observations from monocular cameras. We show that inclusion of visual and identity information aids validation gating for data association when geometric information alone cannot discriminate individual landmarks. The combination of geometric, appearance, and identity information provided a common description (or map) of natural features for each of the nodes in the network. We also show the final map from the live demonstration which includes position estimates and classification labels of the observed features.
Decentralized sensor networks promise virtually unlimited scalability and can tolerate individual component failures. Interaction with human operators plays an important role in the overall system design. The authors describe an experimental active sensor network and identify modes of human-robot interaction that keep up with the network's arbitrary growth.
Humans and robots need to exchange information if the objective is to achieve a task collaboratively. Two questions are considered in this paper: what and when to communicate. To answer these questions, we developed a human–robot communication framework which makes use of common probabilistic robotics representations. The data stored in the representation determines what to communicate, and probabilistic inference mechanisms determine when to communicate. One application domain of the framework is collaborative human–robot decision making: robots use decision theory to select actions based on perceptual information gathered from their sensors and human operators. In this paper, operators are regarded as remotely located, valuable information sources which need to be managed carefully. Robots decide when to query operators using Value-Of-Information theory, i.e. humans are only queried if the expected benefit of their observation exceeds the cost of obtaining it. This can be seen as a mechanism for adjustable autonomy whereby adjustments are triggered at run-time based on the uncertainty in the robots’ beliefs related to their task. This semi-autonomous system is demonstrated using a navigation task and evaluated by a user study. Participants navigated a robot in simulation using the proposed system and via classical teleoperation. Results show that our system has a number of advantages over teleoperation with respect to performance, operator workload, usability, and the users’ perception of the robot. We also show that despite these advantages, teleoperation may still be a preferable driving mode depending on the mission priorities.
This paper describes an information-theoretic approach to decentralised and coordinated control of multi-robot sensor systems. It builds on techniques long established for the related problem of Decentralised Data Fusion (DDF). The DDF architecture uses information measures to communicate state estimates in a network of sensors. For coordinated control of robot sensors, the control objective becomes maximisation of these information measures. A decentralised coordinated control architecture is presented. The approach taken seeks to achieve scalable solutions that maintain consistent probabalistic sensor fusion and payoff formulations. It inherits the many benefits of the DDF method including scalability, seamless handling of sub-system activation and deactivation, and interoperability among heterogeneous units. These features are demonstrated through application to practical multi-feature localisation problems on a team of indoor robots equipped with laser range finders.
Decentralised estimation of heterogeneous sensors is performed on an outdoor network. Attributes such as position, appearance, and identity represented by non-Gaussian distributions are used in in the fusion process. It is shown here that real-time decentralised data fusion of non-Gaussian estimates can be used to build rich environmental maps. Human operators are also used as additional sensors in the network to complement robotic information.
This paper describes an indoor Active Sensor Network, focussing on the implementation aspects of the system, including communication and the application framework. To make the system description more tangible we describe the latest in a series of indoor experiments implemented using ASN. The task is to detect and map motion of people (and robots) in an office space using a network of 12 stationary sensors. The network was operational for several days, with individual platform coming on and off line. On several occasions the network consisted of 39 components. The paper includes a section on the lessons learned during the project’s design and development which may be applicable to other heterogeneous distributed systems with data-intensive algorithms.