HR: 0800h
AN: OS51B-0469    [Abstracts]
TI: Onboard Decision Making For a New Class of AUV Science
AU: * Rajan, K
EM: kanna.rajan@mbari.org
AF: Monterey Bay Aquarium Research Institute, 7700 Sandholdt Road, Moss Landing, CA 95039, United States
AU: McGann, C
EM: cmcgann@mbari.org
AF: Monterey Bay Aquarium Research Institute, 7700 Sandholdt Road, Moss Landing, CA 95039, United States
AU: Py, F
EM: fpy@mbari.org
AF: Monterey Bay Aquarium Research Institute, 7700 Sandholdt Road, Moss Landing, CA 95039, United States
AU: Thomas, H
EM: hthomas@mbari.org
AF: Monterey Bay Aquarium Research Institute, 7700 Sandholdt Road, Moss Landing, CA 95039, United States
AU: Henthorn, R
EM: henthorn@mbari.org
AF: Monterey Bay Aquarium Research Institute, 7700 Sandholdt Road, Moss Landing, CA 95039, United States
AU: McEwen, R
EM: rob@mbari.org
AF: Monterey Bay Aquarium Research Institute, 7700 Sandholdt Road, Moss Landing, CA 95039, United States
AB: Autonomous Underwater Vehicles (AUVs) are an increasingly important tool for oceanographic research. They routinely and cost effectively sample the water column at depths far beyond what humans are capable of visiting. However, control of these platforms has relied on fixed sequences for execution of pre-planned actions limiting their effectiveness for measuring dynamic and episodic ocean phenomenon. At the Monterey Bay Aquarium Research Institute (MBARI), we are developing an advanced Artificial Intelligence (AI) based control system to enable our AUV's to dynamically adapt to the environment by deliberating in-situ about mission plans while tracking onboard resource consumption, dealing with plan failures by allowing dynamic re-planning and being cognizant of vehicle health and safety in the course of executing science plans. Existing behavior-based approaches require an operator to script plans a priori while anticipating where and how the vehicle will transect the water column. While adequate for current needs to do routine pre-defined transects, it has limited flexibility in dealing with opportunistic science needs, is unable to deal with uncertainty in the oceanic environment and puts undue burden on the mission operators to manage complex interactions between behaviors. Our approach, informed by a decades worth of experience in intelligent control of NASA spacecraft, uses a constraint-based representation to manage mission goals, react to exogenous or endogenous failure conditions, respond to sensory feedback by using AI-based search techniques to sort thru a space of likely responses and picking one which is satisfies the completion of mission goals. The system encapsulates the long-standing notion of a sense-deliberate-act cycle at the heart of a control loop and reflects the goal-oriented nature of control allowing operators to specify abstract mission goals rather than detailed command sequences. To date we have tested T- REX (the Teleo-Reactive Executive) on an MBARI Dorado 21" vehicle with a range of scientific instruments for water-column surveys in Monterey Bay. Results to date are available at http://www.mbari.org/autonomy/TREX/index.htm which are very encouraging. Our year-end goals revolve on mapping unstructured phenomenon such as Ocean Fronts and Thin Layers, which we expect will lead to work in adaptive observatory control and autonomous exploration of hydrothermal vents.
UR: http://www.mbari.org/autonomy/TREX/index.htm
DE: 4572 Upper ocean and mixed layer processes
DE: 4594 Instruments and techniques
SC: Ocean Sciences [OS]
MN: 2007 Fall Meeting