OVERVIEW

The ability of a sensor node to move itself or to otherwise influence its location is critical in sensor networks. Actuated sensor networks will be able to self-configure to adapt to realities of inaccessible terrain, develop a robotic ecology for delivering energy sources to other system elements, and obtain coverage of a larger area. Mobile robots provide an additional link between the sensor networks and the physical environment. On the one hand, they can be sensed, monitored and assisted by the sensor network, particularly in the built environment. On the other hand, and more central to the activities of CENS, the robots are nodes in the network with added potential for mobility and actuation in general. In this technology area we focus on projects which exploit sensor-coordinated actuation to enhance the performance of sensing networks.

OBJECTIVES

Our primary scientific objective is to investigate the use of actuation in sensor networks. We are interested in the development of scalable algorithms for the in-network control of actuators mediated by sensed data. Our activities incorporate actuation at three levels. The first is actuation without mobility in which local actuation is informed by data from the sensor network. We focus on developing mechanisms that allow the spatial focus of attention within the actuator network to change as a function of the sensor data without gross motion of the sensors or actuators. In situ, data collection thus occurs with actuators attached to immobile sensors. The principal research challenges in this area are to increase low-level autonomy in the network and develop coordination algorithms for directed sensing and actuation. The second is actuation for limited range mobility in which directed monitoring uses limited forms of motion within limited range, e.g., by using mechanical tethers. Examples include marine monitoring tanks, as well as acoustic, video and other sensors/actuators mounted on tracks/cables to allow multi-elevation monitoring and experimentation. This requires significant new effort in the design and development of tethered robots and their programming before they are ready for deployment. The third is actuation for mobility. Clearly, unconstrained mobility will be needed to apply sensing and actuation across a wider space than fixed infrastructure can feasibly cover; this requires autonomous mobile robots. An example is the use of robotic nodes to fill in gaps in remote monitoring locations, to supply energy to static nodes, or to serve as data-mules. The experimental focus here is on the coordination and control algorithms for autonomous robots with the intent of ultimately applying these techniques to more fully self-configuring systems.

Sensing and Sampling Algorithms
The sensing uncertainty of a sensor network is fundamental to all distributed sensing applications. Sensing uncertainty is directly affected by several factors. First, the sensing medium characteristics, such as the presence of obstacles, limit the sensing range of deployed sensors. Second, the configuration of the sensors with respect to the phenomenon distribution in the environment influences the fidelity of sensing. It is possible to alleviate the effect of these factors by using mobility to align the configuration of the sensing resources to match the medium constraints on sensing and the distribution of an environmental phenomenon over space and time. We have developed an information-theoretic framework to guide this reconfiguration and validate it in a laboratory setting. In information rich natural settings (e.g., Aquatic, Terrestrial), phenomena of interest are inherently multiscale. The main objective of multiscale sensing is the accurate construction of distributed environmental phenomena, such as the temperature over an area, solar illumination near the ground, or the water vapor density in the air. Due to rapid changes of the environment in space and time, it may be prohibitively expensive to achieve this objective at one sensing scale (e.g., by overdeploying sensors. In such situations, we have shown that mobility-enabled multiscale sensing helps in reducing the required resources while meeting or even exceeding the performance of single-scale sensing. Mobility offers the prospect of high-fidelity spatiotemporal sampling. Adaptive sampling techniques with a sound statistical basis are needed to characterize environmental phenomena, and to control mobility. We have explored a suite of adaptive sampling algorithms ranging from event-aware sampling (NIMS), stochastic estimation via particle filters (Aquatic), and biologically inspired taxis-like behaviors for mobile nodes (Aquatic). Specific achievements include a decentralized algorithm that deploys sensor nodes proportional to a given scalar field, locating gradient sources and tracking them over time and boundary finding.

Coordinated Actuation Systems
A fundamental problem is the ability of mobile sensors networks to self-deploy and self-configure. A key challenge in achieving this is that desired network properties are typically global in nature (e.g., degree of connectivity of a network) while the nodes can only sense and act locally. We have made progress on understanding local properties that control global network topology and using these to design control laws for self-deployment of mobile networks. Tiered architectures for large-scale sensor networks include nodes with varying degrees of bandwidth, connectivity, and energy self-sufficiency. Mobile robotic nodes can assist static nodes that can traverse the network, collecting data from the nodes. We have shown experimentally (on land and underwater) that such datamules remove in-network relaying overhead and increase the network lifetime. In the aquatic context there is a need to provide a platform that enables wide-ranging, yet also intensive monitoring and sampling. We have just completed the first prototype of such a platform composed of a robotic boat capable of autonomous navigation and sampling coupled with a distributed network of stationary sensing buoys. (See Aquatic Microbial Observing Systems for more details) Finally, we have made progress on tabletop small mobility platforms for gaming and research support. The robomote platform was designed to address the essential need for tabletop platforms that enable the study of algorithms for mobile sensing. The real action gaming robots (Ragobots) is a laboratory scale test-bed for exploring actuation-related research issues in an engaging setting for students.

The field NIMS system, deployed at the James Reserve, and applied to understory light mapping, microclimate, and phenology, is shown at right.

The NIMS Subsystem
Monitoring of environmental phenomena with embedded networked sensing confronts the challenges of unpredictable variability in the spatial distribution of phenomena, coupled with demands for a high spatial sampling rate in three dimensions. Networked Infomechanical System (NIMS) combines autonomousarticulated and static sensor nodes enabling sufficient spatiotemporal sampling density over large transects to meet a general set of environmental mapping demands. This report discusses several aspects of NIMS (testbed, deployment, task allocation, data gathering, and adaptive sampling). We have augmented the existing NIMS mobile system with a static sensor network and use network-based task allocation for the problem of spatiotemporal monitoring. 8 Figure 1. The field NIMS system, deployed at the James Reserve, and applied to understory light mapping, microclimate, and phenology, is shown at right. Finally, we have built, tested, and deployed several experimental systems. One NIMS system has been developed through several versions and is in constant usage at the James Reserve. A NIMS-laboratory system has been designed and built for laboratory usage supporting experiments in task allocation and sensing uncertainty. A companion NIMS rapid deployment system is applied to water sensing in urban watershed streams.