Engineering topics
Publications and source records attributed to Tang, Benyang.
In this on-going work, we report our efforts of building App Store at OpenNEX, which aims to provide a data analytics software search engine to help researchers find reusable software components to facilitate the development of their own algorithms.
Climate Model Diagnostic Analyzer (CMDA) is a web-based information system designed for the climate modeling and model analysis community to analyze climate data from models and observations. CMDA provides tools to diagnostically analyze climate data for model validation and improvement, and to systematically manage analysis provenance for sharing results with other investigators. CMDA utilizes cloud computing resources, multi-threading computing, machine-learning algorithms, web service technologies, and provenance-supporting technologies to address technical challenges that the Earth science modeling and model analysis community faces in evaluating and diagnosing climate models. As CMDA technology and infrastructure have matured, we have developed the educational and scientific applications of CMDA. Educationally, CMDA supported the summer school of the JPL Center for Climate Sciences in 2014, 2015, and 2016. In the summer school, the students work on group research projects where CMDA provide datasets, analysis tools, and provenance support utility tools. Each student is assigned to a virtual machine with CMDA installed in Amazon Web Services. Scientifically, we have developed several science use cases of CMDA covering various topics, datasets, and analysis types. Each of the science use cases is described in terms of a scientific goal, datasets used, the analysis tools used, scientific results discovered, an analysis result such as output plots and data files, and a link to the corresponding analysis service call with all the input arguments filled.
The comprehensive and innovative evaluation of climate models with newly available global observations is critically needed for the improvement of climate model current-state representation and future-state predictability. A climate model diagnostic evaluation process requires physics-based multi-variable analyses that typically involve large-volume and heterogeneous datasets, making them both computation- and data-intensive. With an exploratory nature of climate data analyses and an explosive growth of datasets and service tools, scientists are struggling to keep track of their datasets, tools, and execution/study history, let alone sharing them with others. In response, we have developed a cloud-enabled, provenance-supported, web-service system called Climate Model Diagnostic Analyzer (CMDA). CMDA enables the physics-based, multivariable model performance evaluations and diagnoses through the comprehensive and synergistic use of multiple observational data, reanalysis data, and model outputs. At the same time, CMDA provides a crowd-sourcing space where scientists can organize their work efficiently and share their work with others. CMDA is empowered by many current state-of-the-art software packages in web service, provenance, and semantic search.
Based only on its appearance in imagery, this program uses close-range 3D terrain analysis to produce training data sufficient to estimate the traversability of terrain beyond 3D sensing range. This approach is called learning from stereo (LFS). In effect, the software transfers knowledge from middle distances, where 3D geometry provides training cues, into the far field where only appearance is available. This is a viable approach because the same obstacle classes, and sometimes the same obstacles, are typically present in the mid-field and the farfield. Learning thus extends the effective look-ahead distance of the sensors.
The Autonomous Exploration for Gathering Increased Science System (AEGIS) provides automated targeting for remote sensing instruments on the Mars Exploration Rover (MER) mission, which at the time of this reporting has had two rovers exploring the surface of Mars (see figure). Currently, targets for rover remote-sensing instruments must be selected manually based on imagery already on the ground with the operations team. AEGIS enables the rover flight software to analyze imagery onboard in order to autonomously select and sequence targeted remote-sensing observations in an opportunistic fashion. In particular, this technology will be used to automatically acquire sub-framed, high-resolution, targeted images taken with the MER panoramic cameras. This software provides: 1) Automatic detection of terrain features in rover camera images, 2) Feature extraction for detected terrain targets, 3) Prioritization of terrain targets based on a scientist target feature set, and 4) Automated re-targeting of rover remote-sensing instruments at the highest priority target.
A software program provides a Sensorweb architecture for alert-processing, event detection, asset allocation and planning, and visualization. It automatically tasks and re-tasks various types of assets such as satellites and robotic vehicles in response to alerts (fire, weather) extracted from various data sources, including low-level Webcam data. JPL has adapted cons iderable Sensorweb infrastructure that had been previously applied to NASA Earth Science applications. This NASA Earth Science Sensorweb has been in operational use since 2003, and has proven reliability of the Sensorweb technologies for robust event detection and autonomous response using space and ground assets. Unique features of the software include flexibility to a range of detection and tasking methods including those that require aggregation of data over spatial and temporal ranges, generality of the response structure to represent and implement a range of response campaigns, and the ability to respond rapidly.
This paper describes technology to support a new paradigm of space science campaigns. These campaigns enable opportunistic science observations to be autonomously coordinated between multiple spacecraft. Coordinated spacecraft can consist of multiple orbiters, landers, rovers, or other in-situ vehicles (such as an aerobot). In this paradigm, opportunistic science detections can be cued by any of these assets where additional spacecraft are requested to take further observations characterizing the identified event or surface feature. Such coordination will enable a number of science campaigns not possible with present spacecraft technology. Examples from Mars include enabling rapid data collection from multiple craft on dynamic events such as new Mars dark slope streaks, dust-devils or trace gases. Technology to support the identification of opportunistic science events and/or the re-tasking of a spacecraft to take new measurements of the event is already in place on several individual missions such as the Mars Exploration Rover (MER) Mission and the Earth Observing One (EO1) Mission. This technology includes onboard data analysis techniques as well as capabilities for planning and scheduling. This paper describes how these techniques can be cue and coordinate multiple spacecraft in observing the same science event from their different vantage points.
Spacecraft processors and memory are subjected to high radiation doses and therefore employ radiation-hardened components. However, these components are orders of magnitude more expensive than typical desktop components, and they lag years behind in terms of speed and size. We have integrated algorithm-based fault tolerance (ABFT) methods into onboard data analysis algorithms to detect radiation-induced errors, which ultimately may permit the use of spacecraft memory that need not be fully hardened, reducing cost and increasing capability at the same time. We have also developed a lightweight software radiation simulator, BITFLIPS, that permits evaluation of error detection strategies in a controlled fashion, including the specification of the radiation rate and selective exposure of individual data structures. Using BITFLIPS, we evaluated our error detection methods when using a support vector machine to analyze data collected by the Mars Odyssey spacecraft. We found ABFT error detection for matrix multiplication is very successful, while error detection for Gaussian kernel computation still has room for improvement.
We discuss a current, ongoing demonstration of insitu onboard detection in which the Earth Observing-1 spacecraft detects surface sulfur deposits that originate from underlying springs by distinguishing the sulfur from the ice-rich glacial background, a good analogue for the Europan surface. In this paper, we describe the process of developing the onboard classifier for detecting the presence of sulfur in a hyperspectral scene, including the use of a training/testing set that is not exhaustively labeled, i.e.not all true positives are marked, and the selection of 12, out of 242, Hyperion instrument wavelength bands to use in the onboard detector. This study aims to demonstrate the potential for future missions to capture short-lived science events, make decisions onboard, identify high priority data for downlink and perform onboard change detection. In the future, such capability could help maximize the science return of downlink bandwidth-limited missions, addressing a significant constraint in all deep-space missions.
An improved algorithm is developed based on support vector regression (SVR) to estimate horizonal water vapor transport integrated through the depth of the atmosphere ((Theta)) over the global ocean from observations of surface wind-stress vector by QuikSCAT, cloud drift wind vector derived from the Multi-angle Imaging SpectroRadiometer (MISR) and geostationary satellites, and precipitable water from the Special Sensor Microwave/Imager (SSM/I). The statistical relation is established between the input parameters (the surface wind stress, the 850 mb wind, the precipitable water, time and location) and the target data ((Theta) calculated from rawinsondes and reanalysis of numerical weather prediction model). The results are validated with independent daily rawinsonde observations, monthly mean reanalysis data, and through regional water balance. This study clearly demonstrates the improvement of (Theta) derived from satellite data using SVR over previous data sets based on linear regression and neural network. The SVR methodology reduces both mean bias and standard deviation comparedwith rawinsonde observations. It agrees better with observations from synoptic to seasonal time scales, and compare more favorably with the reanalysis data on seasonal variations. Only the SVR result can achieve the water balance over South America. The rationale of the advantage by SVR method and the impact of adding the upper level wind will also be discussed.
There are well-established methods for reducing the number of support vectors in a trained binary support vector machine, often with minimal impact on accuracy. We show how reduced-set methods can be applied to multiclass SVMs made up of several binary SVMs, with significantly better results than reducing each binary SVM independently. Our approach is based on Burges' approach that constructs each reduced-set vector as the pre-image of a vector in kernel space, but we extend this by recomputing the SVM weights and bias optimally using the original SVM objective function. This leads to greater accuracy for a binary reduced-set SVM, and also allows vectors to be 'shared' between multiple binary SVMs for greater multiclass accuracy with fewer reduced-set vectors. We also propose computing pre-images using differential evolution, which we have found to be more robust than gradient descent alone. We show experimental results on a variety of problems and find that this new approach is consistently better than previous multiclass reduced-set methods, sometimes with a dramatic difference.
Robotic ground vehicles for outdoor applications have achieved some remarkable successes, notably in autonomous highway following (Dickmanns, 1987), planetary exploration (1), and off-road navigation on Earth (1). Nevertheless, major challenges remain to enable reliable, high-speed, autonomous navigation in a wide variety of complex, off-road terrain. 3-D perception of terrain geometry with imaging range sensors is the mainstay of off-road driving systems. However, the stopping distance at high speed exceeds the effective lookahead distance of existing range sensors. Prospects for extending the range of 3-D sensors is strongly limited by sensor physics, eye safety of lasers, and related issues. Range sensor limitations also allow vehicles to enter large cul-de-sacs even at low speed, leading to long detours. Moreover, sensing only terrain geometry fails to reveal mechanical properties of terrain that are critical to assessing its traversability, such as potential for slippage, sinkage, and the degree of compliance of potential obstacles. Rovers in the Mars Exploration Rover (MER) mission have got stuck in sand dunes and experienced significant downhill slippage in the vicinity of large rock hazards. Earth-based off-road robots today have very limited ability to discriminate traversable vegetation from non-traversable vegetation or rough ground. It is impossible today to preprogram a system with knowledge of these properties for all types of terrain and weather conditions that might be encountered.
Autonomous off-road navigation of robotic ground vehicles has important applications on Earth and in space exploration. Progress in this domain has been retarded by the limited lookahead range of 3-D sensors and by the difficulty of preprogramming systems to understand the traversability of the wide variety of terrain they can encounter.
We are developing a pixel-level cloud-type classifier for the Multi-angle Imaging SpectroRadiometer (MISR), an instrument used to study clouds and aerosols from NASA's Terra satellite.
Autonomous off-road navigation of robotic ground vehicles has important applications on Earth and in space exploration. Progress in this domain has been retarded by the limited lookahead range of 3-D sensors and by the difficulty of preprogramming systems to understand the traversability of the wide variety of terrain they can encounter. Enabling robots to learn from experience may alleviate both of these problems. We define two paradigms for this, learning from 3-D geometry and learning from proprioception, and describe initial instantiations of them we have developed under DARPA and NASA programs. Field test results show promise for learning traversability of vegetated terrain, learning to extend the lookahead range of the vision system, and learning how slip varies with slope.