James Kurien

Spacecraft Autonomy Flight Experience: The DS1 Remote Agent Experiment

In May 1999 state-of-the-art autonomy technology was allowed to assume command and control of the Deep Space One spacecraft during the Remote Agent Experiment. This experiment demonstrated numerous autonomy concepts ranging from high-level goaloriented commanding to on-board planning to robust plan execution to model-based fault protection. Many lessons of value to future enhancements of spacecraft autonomy were learned in preparing for and executing this experiment. This paper describes those lessons and suggests directions of future work in this field.

A hybrid procedural/deductive executive for autonomous spacecraft

The New Millennium Remote Agent (NMRA) will be the first AI system to control an actual spacecraft. The spacecraft domain places a strong premium on autonomy and requires dynamic recoveries and robust concurrent execution, all in the presence of tight real-time deadlines, changing goals, scarce resource constraints, and a wide variety of possible failures. To achieve this level of execution robustness, we have integrated a procedural executive based on generic procedures with a deductive model-based executive. A procedural executive provides sophisticated control constructs such as loops, parallel activity, locks, and synchronization which are used for robust schedule execution, hierarchical task decomposition, and routine configuration management. A deductive executive provides algorithms for sophisticated state inference and optimal failure recovery planning. The integrated executive enables designers to code knowledge via a combination of procedures and declarative models, yielding a rich modeling capability suitable to the challenges of real spacecraft control. The interface between the two executives ensures both that recovery sequences are smoothly merged into high-level schedule execution and that a high degree of reactivity is retained to effectively handle additional failures during recovery.

Costs and Benefits of Model-based Diagnosis

Over the past 20 years, there has been much work in the area of model-based diagnosis (MBD). By this we mean diagnosis systems arising from computer science or artificial intelligence approaches where a generic software engine is developed to address a large class of diagnosis problems. Later, models are created to apply the engine to a specific problem. These techniques are very attractive, suggesting a vision of machines that repair themselves, reduced costs for all kinds of endeavors, spacecraft that continue their missions even when failing, and so on. This promise inspired a broad range of activity, including our involvement over several years in flying the Livingstone and Livingstone 2 on-board model-based diagnosis and recovery systems as experiments on two spacecraft. While a great deal was learned through a variety of applications to simulators, testbeds and flight experiments, no project adopted the technology in operations and the expected benefits have not yet come to fruition. This led us to ask what are the costs of using MBD for the operational scenarios we encountered, what are the benefits, and how do we approach the question of whether the benefits outweigh the costs? How are missions today approaching fault diagnosis and recovery during operations? If we characterize the cost and benefits of using MBD, how would it compare with traditional ways of making a system more robust? How did expectations for MBD compare to benefits seen in the field and why? The literature does provide existing cost models for related endeavors such as integrated vehicle health management. It also provides excellent narratives of why projects chose not to use MBD after considering it. However, we believe that this paper is the first to unpack and discuss the cost, benefit and risk factors that impact the net value of model-based diagnosis and recovery. We use experience with systems such as Livingstone as an example, so our focus is on-board model-based diagnosis and recovery, but we believe many of the insights and remaining questions on the costs and benefits are applicable to other diagnosis applications. Quantitative model of when on-board model-based diagnosis would be an effective choice, it lays out the cost/benefit proposition and identifies several disconnects that we believe prevent adoption as an operational tool. While we do not suggest metrics for every cost, benefit and risk factor we identify, we do discuss where each factor arises in development or operations and how model-based diagnosis and recovery tends to leverage or exacerbate each. As such we believe the analysis is of use to those developing MBD or related techniques and those who may employ them. It also serves as one example of how honest expectations based on technical capability can come to differ from the net impact on customer problems. In this paper we present a cost/benefit analysis for MBD, using expectations and experiences with Livingstone as an example. We provide an overview of common techniques for making spacecraft robust, citing fault protection schemes from recent missions. We lay out the cost, benefit and risk advantages associated with on-board MBD, and use the examples to probe each expected advantage in turn. We conclude our analysis with a summary of our method for analyzing the costs and benefits in a particular domain, and encourage others to come forward with analyses of costs and benefits for fielded systems. Finally, we discuss related work both in terms of similar analyses and fielded systems.

Intrinsic Hurdles in Applying Automated Diagnosis and Recovery to Spacecraft

Experience developing and deploying model-based diagnosis (MBD) and recovery and other model-based technologies on a variety of testbeds and flight experiments led us to explore why our expectations about the impact of MBD on spacecraft operations have not been matched by effective benefits in the field. By MBD, we mean the problem of observing a mechanical, software, or other system and determining what failures its internal components have suffered using a generic inference algorithm and a model of the systems components and interconnections. These techniques are very attractive, suggesting a vision of machines that repair themselves, reduced costs for all kinds of endeavors, spacecraft that continue their missions even when failing, and so on. This promise inspired a broad range of activities, including our involvement over several years in flying the Livingstone and L2 onboard MBD and recovery systems as experiments on Deep Space 1 and Earth Observer 1 spacecraft. Yet, in the end, no spacecraft project adopted the technology in operations nor flew additional flight experiments. To our knowledge, no spacecraft project has adopted any other MBD technology in operations. In this paper, we present a cost/benefit analysis for MBD using expectations and experiences with Livingstone as an example. We provide an overview of common techniques for making spacecraft robust, citing fault protection schemes from recent missions. We lay out the cost, benefit, and risk advantages associated with onboard MBD and use the examples to probe each expected advantage in turn. We suggest a method for evaluating a mission that has already been flown and providing a rough estimate of the maximum value that a perfect onboard diagnosis and recovery system would have provided. By unpacking the events that must occur in order to provide value, we also identify the factors needed to compute the expected value that would be provided by a real diagnosis and recovery system. We then discuss the expected value we would estimate that such a system would have had for the Mars Exploration Rover mission. This has allowed us to identify the specific assumptions that made our expectations for MBD in this domain incorrect.