This project links for the first time high level planning and reconfiguration to low level execution in heterogeneous multi-agent systems. Planing decisions are automated and combine the diverse capabilities of the underlying systems towards a common objective. The derived plans have guaranteed refinement into timed switching sequences of novel, decentralized continuous controllers. The interconnected system can complete a variety of complex cooperative tasks without hardware reconfiguration. This is the first scalable methodology for complete planning and controlled execution in heterogeneous interconnected systems.
The method builds on a hierarchical hybrid architecture in which decision making takes place in the higher purely discrete layers, while the execution of plans is supervised by continuous controllers at the lower layers. The link between the low continuous dynamics and the higher discrete models is established using abstraction. Different homogeneous robotic groups obtain asymptotic abstract discrete representations via a palette of cooperative controllers. Controller switching within a group, as well as concurrent and synchronized execution among several groups is captured in a product-timed automaton. The timed automaton is further abstracted into a purely discrete push down automaton that uses its stack to store timing information. From the push down automaton we extract an equivalent context free grammar, and express the cooperative tasks as words in this grammar. Automatic parsing yields derivations that correspond to cooperative plans, in the form of sequences of controller activations in the underlying groups.
Being one of the three Hispanic Serving Institutions in Carnegie Doctoral/Research Universities-Extensive, UNM is uniquely positioned to recruit Hispanic and Native American students
