Closing the Loop: Control and Robot Navigation in Wireless Sensor Networks

Shawn Michael Schaffert

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2006-112
September 5, 2006

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2006/EECS-2006-112.pdf

Wireless sensor networks have received considerable attention for their potential as a cheap, easily deployed, distributed monitoring tool. Recently, researchers have begun to investigate the use of wireless sensor networks to drive closed-loop control systems. However, such composite systems are nontrivial to design due to the system interface dichotomy: control systems typically assume periodic, high frequency sensor updates whereas sensor networks provide aperiodic, low frequency, and laggy sensor updates. Utilizing robot navigation and pursuit-evasion games as benchmarks, our research focuses on improving control system performance by exploiting the properties of wireless sensor networks. We developed and deployed a real-world, medium-scale wireless sensor network for playing pursuit-evasion games. Using our experience from this deployment, we highlight the difficulties in using sensor network data to accurately localize robots. Several techniques designed to compensate for such difficulties are developed and incorporated into an unified system architecture. To test our architecture, an application-level simulator, accounting for many of the sensor network characteristics that frustrate control design, is developed. This simulator allows us to identify components of our system architecture that can improve the performance of control systems operating in networks of sensors. Amongst the components, intelligent path planning is identified as uniquely important in improving robot localization accuracy during navigation. Path planning techniques that use information maps, exploiting the knowledge of node topology and sensor models, are developed. Information is a metric for measuring the ability of a region in the environment to aid in robot localization. In particular, for each region in the environment, an information map computes the change in entropy expected by a robot in this area using Markov localization. We adapt sensor network models for use with information maps and verify the ability of such maps to improve robot localization. Additionally, automatic path planning techniques based on information maps are developed that minimize localization error. We compare the performance of these planners with other path planners, and demonstrate that this technique is effective for generating paths that increase the accuracy of localization while typically requiring fewer detections.

Advisor: S. Shankar Sastry


BibTeX citation:

@phdthesis{Schaffert:EECS-2006-112,
    Author = {Schaffert, Shawn Michael},
    Title = {Closing the Loop: Control and Robot Navigation in Wireless Sensor Networks},
    School = {EECS Department, University of California, Berkeley},
    Year = {2006},
    Month = {Sep},
    URL = {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2006/EECS-2006-112.html},
    Number = {UCB/EECS-2006-112},
    Abstract = {Wireless sensor networks have received considerable attention for their potential as a cheap, easily deployed, distributed monitoring tool. Recently, researchers have begun to investigate the use of wireless sensor networks to drive closed-loop control systems. However, such composite systems are nontrivial to design due to the system interface dichotomy: control systems typically assume periodic, high frequency sensor updates whereas sensor networks provide aperiodic, low frequency, and laggy sensor updates. Utilizing robot navigation and pursuit-evasion games as benchmarks, our research focuses on improving control system performance by exploiting the properties of wireless sensor networks.
We developed and deployed a real-world, medium-scale wireless sensor network for playing pursuit-evasion games. Using our experience from this deployment, we highlight the difficulties in using sensor network data to accurately  localize robots. Several techniques designed to compensate for such difficulties are developed and incorporated into an unified system architecture. To test our architecture, an  application-level simulator, accounting for many of the sensor network characteristics that frustrate control design, is developed. This simulator allows us to identify components of our system architecture that can improve the performance of control systems operating in networks of sensors. Amongst the components, intelligent path planning is identified as uniquely important in improving robot localization accuracy during navigation.
Path planning techniques that use information maps, exploiting the knowledge of node topology and sensor models, are developed. Information is a metric for measuring the ability of a region in the environment to aid in robot localization. In particular, for each region in the environment, an information map computes the change in entropy expected by a robot in this area using Markov localization. We adapt sensor network models for use with information maps and verify the ability of such maps to improve robot localization. Additionally, automatic path planning techniques based on information maps are developed that minimize localization error. We compare the performance of these planners with other path planners, and demonstrate that this technique is effective for generating paths that increase the accuracy of localization while typically requiring fewer detections.}
}

EndNote citation:

%0 Thesis
%A Schaffert, Shawn Michael
%T Closing the Loop: Control and Robot Navigation in Wireless Sensor Networks
%I EECS Department, University of California, Berkeley
%D 2006
%8 September 5
%@ UCB/EECS-2006-112
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2006/EECS-2006-112.html
%F Schaffert:EECS-2006-112