Mixed Initiative Control for Automa-teams

The MICA project has five major research thrusts: (1) hierarchical architecture design for semi-automated, distributed teams of agents, (2) control of hybrid systems, (3) management of information, (4) confrontation of uncertainty, and (5) incorporating human intervention in mission planning and execution. The overall system architecture features flexible team formation, task specification, pre-mission evaluation, and changes in goals, team composition, and tasks during mission execution. Berkeley's simulation tools, such as SHIFT and Ptolemy will be the initial platforms for design and evaluation. Simulation models will be integrated into the Boeing OEP platform. The hybrid control component is organized in two levels. Low-level controllers govern the real-time behavior of physical resources such as vehicles. Higher-level control is embodied in a hierarchy of semi-autonomous teams tasked with planning, information collection and distribution, assessment of mission execution, and management of teams under their authority. Low-level control will be based on continuous-time control techniques; the design of high-level teams consists of numerical and symbolic algorithms. SHIFT and Ptolemy will serve to codify and evaluate the control design libraries in different mission scenarios. In the information management component, the project will look at patially distributed teams operating in a dynamic environment collecting and exchanging necessary information over an unreliable network. This component will make uses of advances achieved in the ONR-sponsored WoW (Web over Wireless) project and the DARPA sponsored Smart Networks project (part of ITO Program NMS). Three types of uncertainty will be studied in the context of mission planning and execution. Non-deterministic uncertainty will be handled through tools based on the theory of verification. Probabilistic uncertainty will be assessed via Markovian models. Adversarial uncertainty will be conceptualized in game-theoretic approaches. Our control designs will take into account the appropriate type of uncertainty. Mission planning and execution must incorporate three modes of human intervention at all levels of decision-making. First, automatically synthesized plans, task and team allocations, may be presented to a human operator for approval and modification. Second, during mission execution, reports from teams may be forwarded to a human operator who may intervene to abort or modify team tasks. Third, when a team encounters an unanticipated situation, an "exception" is invoked and handled either by a superior agent or a human operator. Such exception-handling capability is essential to any highly automated mission.

This group has the following subpages:

©2002-2018 U.C. Regents