[citations and links to be added soon.]
1984-2004
Spacecraft Control Center Automation
My work (1984-1994) focused on advances in heuristic and model-based reasoning techniques for the support of automated monitoring and diagnosis, the development of artificial intelligence software engineering tools to support applications for monitoring and diagnosis, and the test and evaluation of these techniques in scaled-up applications under realistic or actual operational conditions.
A major contribution of the work was characterization of the automated spacecraft monitoring and diagnosis process as a set of coordinated problem-solving tasks and the investigation of automated reasoning techniques that can support or automate each task. With my colleagues, algorithms were developed for plan analysis, sensor selection, and predictive generation of expected sensor values, including automated dynamic alarm limit adjustment based on abstract models of the spacecraft (Doyle, R., Sellers, S. and Atkinson, D.). Methods for selective monitoring of sensors during device operation were conceived (Doyle, R., Atkinson, D., and Doshi, R.). New model-based and procedure-based diagnosis algorithms for explaining hardware, environmental, and plan anomalies were developed (Lam R., Doshi R., Atkinson, D., and Lawson, D.). Novel AI software engineering tools were created to support the application of these techniques (James, M. and Atkinson, D.). Each new method was developed and tested singly or in combination in several software and hardware prototypes, including robotic systems, and space systems operations control systems.
A second major contribution was the investigation of intelligent diagnosis systems for scaled-up applications that address the complete monitoring and diagnostic function for robotic spacecraft operations (Atkinson, D., Lawson, D, and James, M.). This has required new technology for the architecture of such systems, such as the coordination of hierarchical knowledge-based systems for diagnosis (Atkinson, D. and James, M.). A software prototype for intelligent spacecraft health-monitoring and diagnosis was developed and successfully tested in actual mission operations during the Voyager II spacecraft encounter with Neptune and Uranus, the Magellan spacecraft operations at Venus, and the Galileo spacecraft cruise to Jupiter (Atkinson, D. & James, M.). These have been among the most challenging operational conditions in the planetary exploration program. This research fed directly into advanced development and operational systems for multiple spacecraft, control centers and ground data systems, launch systems.
The “Spacecraft Health Automated Reasoning Prototype” (SHARP) system, was successfully used to monitor and diagnosis Telecommunications Link Status for Voyager II during the Neptune encounter, It was the first to detect and helped to isolate a fault that would have resulted in complete loss of the science data. SHARP was later applied to Magellan and Galileo operations and its technology is now incorporated in JPL’s multi-mission operations control center systems. Now called SHINE and licensed by Caltech, the technology has also been used by Arianne ESPACE, both Boeing and Lockheed in the Joint Strike Fighter Program, and other companies for various applications in Homeland security, the financial and pharmaceutical industries. With M. James.
Software Systems Architectures for Robotic Spacecraft Autonomy
The constraints of space operations (such as harsh environments, or round-trip light time delay) introduce unique requirements for space robotics. This is also a environment where the bar is set very high for both performance and reliability. Numerous Ph.D students finished their dissertations at my lab, or had post-doc positions, and our collective research efforts resulted in numerous algorithms, architectures and demonstrable applications for autonomous space robotics. These included autonomous surface navigation in rough natural terrain with a 100 meter out-and-back demonstration in 1990, one of the first. We developed methods for reasoning about geometrical relations and spatial constraints, and embedded these in a task planning and control system for a robot that performed automated satellite repair (laboratory demonstration). With R. Desai, R. Doshi and others.
Micro-robots for Planetary Surface Exploration
Dr. Colin Angle, a student of Prof. Rodney Brooks at MIT, spent a summer at JPL where he, my colleague and summer faculty visitor Dr. David P. Miller and I applied Brook's theories to the development of behavior-control of micro-robotics in space exploration. Our work of that summer literally changed the course of JPL robotic exploration from enormous vehicles to small, agile robots. The successful development and mission of the Sojourner robot on the Pathfinder mission to Mars was a direct consequence of the branch of robotic research I initiated at JPL. Angle, Brooks and others later founded iRobot; the Roomba robot has among its roots the work conducted that summer in my lab.
Machine Learning and Automation for Scientific Data Analysis
An early contribution of my research team has been one of architecting a trainable learning system that can efficiently process nearly a terabyte of sky survey data, detect and classify galaxies at a faster and more effective rate than humans. In this project, I supervised the research of Dr. Usama Fayyad in creating a system that was successfully able to create a complete catalog of galaxies from multidimensional data in the Mount Palomar Digital All-Sky Survey, a project led by Prof. Stan Djorgovski at Caltech. A human team would have taken many years - possibly a decade – to complete the catalog. The SKICAT system completed it in fewer than four months of continuous processing (U. Fayyad, P. Smyth and D. Atkinson). This accomplishment has been applied and extended by others to automated recognition and classification of craters in planetary imaging, classification of solar phenomenon, small body detection, and other studies. The success of these endeavors lead directly to the space science initiative called “Virtual Observatory”.
I envisioned combining the data analysis and classification technology with the autonomous control architectures we had been developing. I proposed a project to pursue this idea (“Science Analysis Associate”), and investigated some of the foundations and requirements for an integrated system. This idea of melding spacecraft control automation with science analysis was an inspiration for the research program of one of my colleagues, Dr. Steve Chien who has carried it forward with original research and technical maturation such that the resulting systems are today in operational, autonomous control of the EO-1 spacecraft and also used for autonomous science operations on the Mars planetary rovers.
Integrated Computing, Simulation, Modeling and Visualization (CSMV)
Lead Investigator for CMSV project to create software environment and resusable applications to aid science and engineering at Caltech/JPL. The CSMV project approach was to “design for, and enable re-use”, which the research team believed was extremely important for success in the entire lifecycle of research, development, and ultimately the dissemination and use of our technology. The project investigated, created and demonstrated middleware, applications and standards that enabled engineers and scientists to easily find the models they needed in a widely distributed computing environment, to determine what the model does, and to enable the automated, interoperable use of these models in engineering or science applications. A key accomplishment was a highly distributed library of reusable tools for modeling, simulation and visualization. The computational framework enabled interoperability of domain-specific models, tools and parameters based on standardized metadata, distributed model service protocols, and model exchange service mechanisms. Demonstrated and delivered to projects modeling sunspot and flare activity, Mars planetology and rapid visualization of rover photo mosaics captured on Mars surface. With Meemong Lee, Mark Kordon, Erik De Jong, Kevin Hussey and Dan Crichton.