CPS: Creative Problem Solver

Military systems operate in complex, adversarial environments where their initial planned operations are often overcome by events, but soldiers use those systems in creative ways to achieve their mission objectives. For example, complex ground robotic systems that were initially deployed for surveillance and explosive ordinance disposal have instead been used as decoys, ammunition carriers, and other creative applications. While AI systems are creating a revolution in civilian autonomous systems, they are not aimed at achieving this type of creativity in the face of adversity. Generating creative plans requires a more fundamental understanding of the capabilities and potential functions of system components. We propose to develop CPS, a planning and plan execution system that will enable future autonomous systems to creatively solve problems for which they were not originally designed or prepared, yielding robust autonomous systems that achieve mission success even in novel, unexpected scenarios. Key research advances will include: -Abstract models for understanding how system components can be combined in novel ways. -Creative configuration and hybrid systems planning to create new compositions of primitive capabilities that will achieve complex system goals. -Adaptation via trial and error, robustly achieving goals despite inaccurate models and continuously degrading capabilities.