Implementing fully autonomous space missions depends on an agent's ability to successfully integrate the symbolic and geometric deliberation required to plan and achieve action. In the domain of space-based intelligent self-assembly and reconfiguration (ISAR), we find one of the best arguments for the execution of intelligent motion due to the extra burden of spacecraft coordinating complicated real-time maneuvers in parallel to achieve an aggregate goal. By partitioning this domain into different operational phases, broadly studying related research/applications, and extracting/consolidating relevant attributes particular to a phase, we find new heuristics that help agents self-regulate individual and group-based priorities and strategies, independent of spacecraft size. Whether to throttle movement by limiting thruster burns in more dense neighborhoods, synchronize the relative motion of many agents simultaneously, or change velocity, etc. depends on collective group goals, dynamic constraints and mitigating between fuel/time losses and science objectives. We have developed a tightly integrated task and trajectory planning architecture in C++ and a common symbolic framework from which to conduct a variety of ISAR simulations. The modularity of this architecture allows us to use different trajectory planners-e.g., we use an astrodynamics expert based on Lambert's method for orbital insertions and another based on the Clohessy and Wiltshire form of Hill's equations for proximity operations-and abstract the computational demands of applying heuristics and satisfying cost functionals within a module called the "translator." This module serves as the interface hub of all planners, passing continuously maintained ISAR component state vectors and cost/heuristic information as needed. The utility of our research is demonstrated and analyzed from the quantitative computational results of simulations conducted, including: Earth-based assembly of a very large orbital reflector telescope and solar power array; dynamic formation flying; and lunar/Mars ground-based scenarios. Though there have been many advances within this domain, these tend to be focused on developing new AI or control techniques. Our on-going research continues to dwell in the middle of these efforts by delving into the practical and algorithmic boundaries of planning and mathematics. We show that a domain ostensibly ruled by mathematics presents compelling new AI components/essential heuristics from which to fashion integrated symbolic and geometric reasoning that is applicable to a variety of autonomous space-based missions.
展开▼
