Title of Presentation: Improving the Science Return from Coastal Sensor Webs Using Autonomous Predictive Control and Resource Management
Primary (Corresponding) Author: Ashit Talukder
Organization of Primary Author: Jet Propulsion Lab
Co-Authors: Alan Blumberg, Anand Panangadan, Thomas Herrington, Nickitas Georgas
Abstract: We discuss a general technique for resource management in heterogeneous sensor webs for environmental monitoring and other applications. In this AIST-funded project, a novel Model Predictive Control (MPC) framework to adaptively control real-time resources of a variety of static & mobile sensor web assets with limited resources was developed and shown to improve the performance of the New York Harbor Prediction and Observation Sensor Web System (NYHOPS).
The New York Harbor Prediction and Observation Sensor Web System (NYHOPS) has been in operation for several years and comprises of a network of sensors to monitor coastal and ocean parameters (elevation, salinity, temperature, and water velocity) in the densely populated regions of the Hudson-Raritan Estuary and the New Jersey Atlantic Ocean shoreline. The current NYHOPS sensor web operates in a fixed mode of operation without any adaptation, which is severely limiting for observing and modeling many kinds of spatiotemporal events such as plumes, storm surges.
MPC assumes a control ìmodel-basedî solution for sensor web adaptation. Determination of the optimal control policy is posed as an optimization problem where the function to be optimized represents the relative utility and cost of performing the control actions. The constraints represent the physical limitations of the system such as energy and wireless network bandwidth. The optimal control policy is obtained by constrained minimization of the objective function at every control step. The predictive capability of the MPC enables a ìlook-aheadî capability that considers future states/events to make the current optimal control decision.
In CARDS, we used the MPC to adapt the overall NYHOPS system, including the sampling rates of static sensors, communication bandwidth, dynamic re-assignment of sensors to relay nodes, and the movement of mobile underwater unmanned vehicles in response to observed events, and thereby improve the accuracy and predictive capability of the ECOMSED model for coastal events including plumes and storm surges. Optimal control of the motions of multiple UUVs guaranteed that the UUV sensor observations had a maximal impact on increasing the predictive power of the ECOMSED model while ensuring that inordinate amounts of energy were not expended in UUV motion. The CARDS sensor web adaptive control solution has been found to improve the accuracy of event modeling and prediction in NYHOPS by 50%. We have developed sensor web visualization tools in Google Earth for use by scientists and customers of the NYHOPS data. CARDS is directly applicable to a variety of sensor webs and paves the way for coordinating multiple ground and space assets owned by NASA for faster and better detection, tracking and characterization of earth and extra-terrestrial environments.