Intelligent and Learning Autonomous Systems: Composability and Correctness
Abstract
The proposed research program will develop new science for synthesis (i.e. design, operation,performance evaluation) of intelligent"" and learning autonomous systems, with assuredcomposability and correctness, even when they interact with unknown and unstructured"" physicalenvironments necessitating on-line learning. We will develop novel system and componentmodels, novel system integration m""ethods, efficient algorithms, implementations,experimentation, focusing on:(i) Development and evaluation of novel formal methods"" that support composability and canachieve useful results for defined environmental interactions at design-time, and rigorouslybal""ance design-time verification with real-time, on-line methods.(ii) Development and evaluation of efficient methods to learn and sol"ve problems safely as theautonomous system encounters new situations in the field and even bound a systemcomponent when it exceeds" its area of competence, training, or knowledge. We willinclude methods and algorithms for sensory-based perception, cognition and" reasoning.(iii)Development of novel and efficient methods to verify and assure composability and correctsystem behavior. We will" develop methods to assure acceptable operations during learningand problem solving, system recovery, with practical useful complex""ity and time response.The development of robust, autonomous systems that interact correctly with unstructuredphysical environments" requires mechanisms in the system for specifying environmentalphenomena that the system is intended to react to; for verifying that the system indeed respondscorrectly to these anticipated scenarios; for detecting new phenomena as they arise during systemopera"tion and responding safely in the moment; and for adapting system specifications, designsand correctness arguments to ensure optima""l system behavior in the future. We propose todevelop techniques that combine the use of model-based systems engineering, design co""ntracts,run-time monitoring and specification learning to facilitate this vision of autonomous-systemcorrectness.The research tea""m brings together three accomplished senior investigators from theUniversity of Maryland College Park, with extensive multi-discipl"inary expertise and worldrenowned accomplishments in the diverse fields needed for the proposed research program. Eachsenior inves"tigator will be actively involved in at least two of the eight tasks, and all willcontribute to the integration, testing and valida""tion work. In addition one postdoctoral fellow, sixgraduate and five undergraduate students will participate in the research each y""ear. We will usethe extensive laboratory and testbed facilities of two world renowned Institutes of the Universityof Maryland, the"" Institute for Systems Research and the University of Maryland Institute forAdvanced Computer Studies, for testing and experimental"" validation of the new synthesis anddesign methods. The program includes a substantial education and training component, as wellas" knowledge dissemination and technology transition efforts.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 09, 2017
- Source ID
- N000141712622
Entities
People
- John Baras
Organizations
- Office of Naval Research
- United States Navy
- University of Maryland