1,721,002 research outputs found
An information-theoretic approach to data fusion and sensor management
Full text linkThe use of multi-sensor systems entails a Data Fusion and Sensor Management requirement in order to optimize the use of resources and allow the synergistic operation of sensors. To date, data fusion and sensor management have largely been dealt with separately and primarily for centralized and hierarchical systems. Although work has recently been done in distributed and decentralized data fusion, very little of it has addressed sensor management. In decentralized systems, a consistent and coherent approach is essential and the ad hoc methods used in other systems become unsatisfactory. This thesis concerns the development of a unified approach to data fusion and sensor management in multi-sensor systems in general and decentralized systems in particular, within a single consistent information-theoretic framework. Our approach is based on considering information and its gain as the main goal of multi-sensor systems. We develop a probabilistic information update paradigm from which we derive directly architectures and algorithms for decentralized data fusion and, most importantly, address sensor management. Presented with several alternatives, the question of how to make decisions leading to the best sensing configuration or actions, defines the management problem. We discuss the issues in decentralized decision making and present a normative method for decentralized sensor management based on information as expected utility. We discuss several ways of realizing the solution culminating in an iterative method akin to bargaining for a general decentralized system. Underlying this is the need for a good sensor model detailing a sensor's physical operation and the phenomenological nature of measurements vis-a-vis the probabilistic information the sensor provides. Also, implicit in a sensor management problem is the existence of several sensing alternatives such as those provided by agile or multi-mode sensors. With our application in mind, we detail such a sensor model for a novel Tracking Sonar with precisely these capabilities making it ideal for managed data fusion. As an application, we consider vehicle navigation, specifically localization and map-building. Implementation is on the OxNav vehicle (JTR) which we are currently developing. The results show, firstly, how with managed data fusion, localization is greatly speeded up compared to previous published work and secondly, how synergistic operation such as sensor-feature assignments, hand-off and cueing can be realised decentrally. This implementation provides new ways of addressing vehicle navigation, while the theoretical results are applicable to a variety of multi-sensing problems
A Bayesian approach to optimal sensor placement
Full text linkBy "intelligently" locating a sensor with respect to its environment it is possible to minimize the number of sensing operations required to perform many tasks. This is particularly important for sensing media which provide only "sparse" data, such as tactile sensors and sonar. In this thesis, a system is described which uses the principles of statistical decision theory to determine the optimal sensing locations to perform recognition and localization operations. The system uses a Bayesian approach to utilize any prior object information (including object models or previously-acquired sensory data) in choosing the sensing locations
A Natural Feature Representation for Unstructured Environments
No full textThis paper addresses the long-standing problem of feature representation in the natural world for autonomous navigation systems. The proposed representation combines Isomap, which is a nonlinear manifold learning algorithm, with expectation maximization, which is a statistical learning scheme. The representation is computed off-line and results in a compact, nonlinear, non-Gaussian sensor likelihood model. This model can be easily integrated into estimation algorithms for navigation and tracking. The compactness of the model makes it especially attractive for deployment in decentralized sensor networks. Real sensory data from unstructured terrestrial and underwater environments are used to demonstrate the versatility of the computed likelihood model. The experimental results show that this approach can provide consistent models of natural environments to facilitate complex visual tracking and data-association problems
Coupled dynamical system based hand-arm grasp planning under real-time perturbations
Full text linkRobustness to perturbation has been advocated as a key element to robot control and efforts in that direction are numerous. While in essence these approaches aim at "endowing robots with a flexibility similar to that displayed by humans", few have actually looked at how humans react in the face of fast perturbations. We recorded the kinematic data from human subjects during grasping motions under very fast perturbations. Results show a strong coupling between the reach and grasp components of the task that enables rapid adaptation of the fingers in coordination with the hand posture when the target object is perturbed. We develop a robot controller based on Coupled Dynamical Systems that exploits coupling between two dynamical systems driving the hand and finger motions. This offers a compact encoding for a variety of reach and grasp motions that adapts on-the-fly to perturbations without the need for any re-planning. To validate the model we control the motion of the iCub robot when reaching for different objects.LAS
Exploiting Variable Stiffness in Explosive Movement Tasks
No full textIt is widely recognised that compliant actuation is advantageous to robot control once high-performance, explosive tasks, such as throwing, hitting or jumping are considered. However, the benefit of intrinsic compliance comes with high control complexity. Specifically, coordinating the motion of the system through a compliant actuator and finding a task-specific impedance profile that leads to better performance is non-trivial. Here, we utilise optimal control to devise time-varying torque and stiffness profiles for highly dynamic movements in compliantly actuated robots. The proposed methodology is applied to a ball-throwing task where we demonstrate that: (i) the method is able to tailor impedance strategies to specific task objectives and system dynamics, (ii) the ability to vary stiffness leads to better performance in this class of movements, (iii) in systems with variable physical compliance, our methodology is able to exploit the energy storage capabilities of the actuators. We illustrate these in several numerical simulations, and in hardware experiments on a device with variable physical stiffness
Fast re-parameterisation of gaussian mixture models for robotics applications
No full textAutonomous navigation and picture compilation tasks require robust feature descriptions or models. Given the non-Gaussian nature of sensor observations, it will be shown that Gaussian mixture models provide a general probabilistic representation allowing analytical solutions to the update and prediction operations in the general Bayesian ltering problem. Each operation in the Bayesian lter for Gaussian mixture models multiplicatively increases the number of parameters in the representation leading to the need for a re-parameterisation step. A computationally ecient re-parameterisation step will be demonstrated resulting in a compact and accurate estimate of the true distribution
A Bayesian approach for place recognition
No full textThis paper presents a robust place recognition algorithm for mobile robots. The framework proposed combines nonlinear dimensionality reduction, nonlinear regression under noise, and variational Bayesian learning to create consistent probabilistic representations of places from images. These generative models are learnt from a few images and used for multi-class place recognition where classification is computed from a set of feature-vectors. Recognition can be performed in near real-time and accounts for complexity such as changes in illumination, occlusions and blurring. The algorithm was tested with a mobile robot in indoor and outdoor environments with sequences of 1579 and 3820 images respectively. This framework has several potential applications such as map building, autonomous navigation, search-rescue tasks and context recognition
- …
