Search

Large Language Models (LLMs) excel in data synthesis but can be inaccurate in domain-specific tasks, which retrieval-augmented generation (RAG) systems address by leveraging user-provided data. However, RAGs require optimization in both retrieval and generation stages, which can affect output quality. In this paper, we present LLM-Ref, a writing assistant tool that aids researchers in writing articles from multiple source documents with enhanced reference synthesis and handling capabilities. Unlike traditional RAG systems that use chunking and indexing, our tool retrieves and generates content directly from text paragraphs. This method facilitates direct reference extraction from the generated outputs, a feature unique to our tool. Additionally, our tool employs iterative response generation, effectively managing lengthy contexts within the language model's constraints. Compared to baseline RAG-based systems, our approach achieves a $3.25\times$ to $6.26\times$ increase in Ragas score, a comprehensive metric that provides a holistic view of a RAG system's ability to produce accurate, relevant, and contextually appropriate responses. This improvement shows our method enhances the accuracy and contextual relevance of writing assistance tools., Comment: 20 pages, 7 figures

Large language models (LLMs) with billions of parameters and pretrained on massive amounts of data are now capable of near or better than state-of-the-art performance in a variety of downstream natural language processing tasks. Neural machine translation (NMT) is one such task that LLMs have been applied to with great success. However, little research has focused on applying LLMs to the more difficult subset of NMT called simultaneous translation (SimulMT), where translation begins before the entire source context is available to the model. In this paper, we address key challenges facing LLMs fine-tuned for SimulMT, validate classical SimulMT concepts and practices in the context of LLMs, explore adapting LLMs that are fine-tuned for NMT to the task of SimulMT, and introduce Simul-LLM, the first open-source fine-tuning and evaluation pipeline development framework for LLMs focused on SimulMT., Comment: ACL 2024

Synchronized formation rotations are a common maneuver for planned precision formations. In such a rotation, attitudes remain synchronized with relative positions, as if the spacecraft were embedded in a virtual rigid body. Further, since synchronized rotations are needed for science data collection, this maneuver requires the highest precision control of formation positions and attitudes. A recently completed, major technology milestone for the Terrestrial Planet Finder Interferometer is the high-fidelity, ground demonstration of precision synchronized formation rotations. These demonstrations were performed in the Formation Control Testbed (FCT), which is a flight-like, multi-robot formation testbed. The FCT is briefly introduced, and then the synchronized rotation demonstration results are presented. An initial error budget consisting of formation simulations is used to show the connection between ground performance and TPF-I flight performance.

The Pointing History Engine (PHE) is a computer program that provides mathematical transformations needed to reconstruct, from downlinked telemetry data, the attitude of the Spitzer Space Telescope (formerly known as the Space Infrared Telescope Facility) as a function of time. The PHE also serves as an example for development of similar pointing reconstruction software for future space telescopes. The transformations implemented in the PHE take account of the unique geometry of the Spitzer telescope-pointing chain, including all data on relative alignments of components, and all information available from attitude-determination instruments. The PHE makes it possible to coordinate attitude data with observational data acquired at the same time, so that any observed astronomical object can be located for future reference and re-observation. The PHE is implemented as a subroutine used in conjunction with telemetry-formatting services of the Mission Image Processing Laboratory of NASA s Jet Propulsion Laboratory to generate the Boresight Pointing History File (BPHF). The BPHF is an archival database designed to serve as Spitzer s primary astronomical reference documenting where the telescope was pointed at any time during its mission.

A number of international space agencies and organizations, to include the National Aeronautics and Space Administration (NASA), the European Space Agency (ESA), and the Centre National d'Etudes Spatiales (CNES), to name a few, have embraced the concept of spacecraft formation flying to revolutionize the capabilities of astronomy and Earth remote sensing from space. The concept has been around well over a decade and a wide array of technologies and capabilities have been developed to enable multiple spacecraft to collaborate in a highly-coupled manner as would be required for a formation flying mission. Furthermore, many relevant capabilities for formation flying have been demonstrated in the area of rendezvous and docking, loosely-controlled formations, and in missions with collaborating spacecraft with very precise metrology. .However, in considering the case of precision formation flying (PFF), i.e, when the relative geometry of multiple vehicles must be controlled on-board in a continuous and precise manner, there have been several missions proposed, but the realization in space has not yet occurred due to a range of issues. This paper will briefly examine those issues and present a concept for demonstrating a core capability for performing PFF, necessary for virtually any PFF mission concept, that will help to overcome the problems encountered in prior attempts and help to allay the risks to enable future PFF science missions.

In this paper, each subsystem of the Celestial Sensor is discussed with emphasis given to the filtering algorithms. The celestial sensor software processes sequential frame based bearing measurements on a peripheral CPY specifically designed for this application. A frame preprocessor is used to normalize each bearing measurement and apply a number of accept/reject rules. The accepted set of measurements is then passed to an Extended Kalman Filter (EKP) that is tuned to track the motion of the robot within the room.

This paper describes the design and construction of the Formation Control Testbed. The FCT consists of three robots navigating on an air bearing floor, propelled by cold gas thrusters. Each robot contains an attitude platform supported on a spherical air bearing which provides three rotational degrees of freedom. The sixth degree of freedom, vertical translation, will be provided by a powered vertical stage, actively controlled to provide a simulated zero-g environment for the attitude platform.

UNKNOWN

Report describes theoretical and experimental study of robustness of adaptive model-reference control scheme designed to suppress vibrations in large, flexible structure. Model can be mathematical model or collection of models approximating dynamics of structure, vibration sensors, and vibration-suppressing actuators.

Report describes experiments in use of six-input/six-output multivariable adaptive control system to suppress vibrations in complicated flexible structure. Represents significant upgrade relative to two-input/two-output system, yielding significant improvement in spatial controllability and effective damping of largest set of modes. Structure exhibits low modal frequencies and complicated dynamics like those of large structures in outer space. Conclusions drawn from research relevant to such applications as active suppression of wind and earthquake vibrations in tall buildings and other large terrestrial structures.

An experimental study is performed to investigate the robustness of model reference adaptive control for the large flexible structures control application. The main nonidealities of concern are unmodeled dynamics, input saturation, and time delay effects (here, actuator and sensor dynamics are lumped into the last item for convenience). This study focuses on the robustness with respect to input saturation and time delay effects, since robustness to unmodeled dynamics is inherent to the basic algorithm and has been demonstrated experimentally elsewhere.