Overview

The multiscale actuation and sensing area covers four topics:  (1) computational multi-scale sensing and environmental modeling, (2) deployment via iteration and mobility, (3) computational signal processing and algorithms for camera and acoustic networks, and (4) the development of the NIMS platform and its deployment in novel environments.

Computational and Multiscale Sensing: Our goal is to develop a new conceptual approach to computational sensing—an emerging scientific and engineering field defined as the process of extraction, analysis and use of knowledge about the instrumented environment and sensed phenomena. Coordinated development and use of non-parametric statistical models, combinatorial and continuous optimization algorithms, and mathematical logic will provide techniques and tools for creating and evaluating sensing procedures. Our initial targets are rain and wind identification and prediction, hypothesis checking-driven deployment, and soil structure reverse engineering. Perhaps the most challenging scaling problem for distributed sensing is that of large region characterization by distributed sensing. Many applications require both microscopic scale detection of phenomena and wide area measurement to enable regional characterization.  The problem of ecosystem monitoring and the optimization of land use, the distribution of water resources, or the characterization of environmental contaminants are all examples in which the scales of centimeters to kilometers may contribute to investigations.  Uniformly dense sensor deployment is not feasible due to resource constraints, in particular in 3-D. Thus, we are pursuing distributed sensing systems that combine sensing on multiple scales for both multimodal sensor data fusion and guidance of actuated sensing.

Deployment via Iteration and Mobility: Selection of sensing locations is a difficult problem. By using an iterative deployment methodology, a few sensors can be placed and the resulting data can be used to learn about additional deployment locations. We have developed and applied this technique to the problem of sensor placement in soil monitoring applications; we are exploring the use of portable, and noninvasive, sensing arrays (auditioning arrays) to make measurements at prospective deployment sites to quickly gain insight into below ground processes. Multi-objective deployments (e.g., ones that provide a required quality of sensing coverage while minimizing overall communication cost) are a natural application of iteration. In the context of a low-power camera network in a complex, outdoor environment, we have shown how to satisfy coverage and connectivity requirements with this approach. The system finds the smallest set of sensor locations that provide adequate sensing coverage and then suggests where to place a small number of additional radios to decrease overall communication cost. In some sensing applications (e.g., aquatic sampling), measurement systems must cover a large space with limited resources. In addition to iterative deployment, we use robots to convey sensors in a space to achieve the required coverage. Planning the motion of robots to maximize the information collected within the constraints of resource bounds (e.g., path length or energy) is NP-hard.  We have developed an approach to coordinate multiple robots, each having a resource constraint, to maximize the ‘informativeness’ of their visited locations. A Gaussian Process is used to model the underlying phenomenon, and the mutual information between the visited locations and remainder of the space is used to characterize the amount of information collected. This algorithm comes with a strong theoretical approximation guarantee due to the submodularity property of mutual information. We have investigated the case where a single mobile robot is used to augment a static sensing system with the goal of reconstructing a scalar field (e.g., temperature) from samples. Our adaptive sampling algorithm uses local linear regression to plan a path by which the resulting field reconstruction error is minimized subject to the energy constraint on the robot. Data from extensive field experiments with the NAMOS and NIMS systems validate the performance of both planning algorithms.

Camera and Acoustic Networks: A camera is an information-rich sensor. We have studied large-scale distributed sensing via imagers, leading to the most widely-used first generation wireless image platform, Cyclops. Through pilot applications and system studies Cyclops has provided a wealth of data and insight into image network challenges and opportunities. Cyclops has been used extensively by researchers at CENS and elsewhere to (1) investigate the application of computer vision techniques to generate high-quality information from a time-series of relatively low-resolution images, (2) study low-power computational paradigms for vision, (3) explore vision architectures and system issues, and (4) analyze communication vs. computation constraints. Some of the driving applications for this distributed image technology are described in the Terrestrial ecology section.

Like cameras, microphones are also a rich source of information. This year, we verified a newly formulated 3D acoustical array for distributed sensing applications and implemented the previously proven approximate-maximum-likelihood (AML) algorithm for animal acoustic source detection and localization. For instance, we collected and processed extensive field data of various birds in the forests of Mexico and marmots in the meadows of Colorado. In addition, we conducted a field test of a newly formulated cooperative vehicle detection and localization scheme using wireless acoustical arrays.

NIMS Platform Development and Deployments: We modified the existing NIMS RD terrestrial system to suit the requirements of a semi-permanent installation at NEON field sites. Whereas the existing system has been targeted towards rapid deployments at remote sites for up to one week of operation, the new system is targeted towards long-term operation with an emphasis on longevity. Two major terrestrial NIMS RD deployments occurred in 2006, apart from the continuing terrestrial program of NIMS RD over the AMARSS transect at the James Reserve. The first deployment in the White Mountains used an updated NIMS RD system to map the surface temperature of an alpine fellfield over 24 hours with 1 cm resolution in space and 1 minute in time. These data demonstrate the significance of microscale patterns of soil and plant surface temperatures in modeling ecological and ecosystem processes. The second deployment was a one-week study at the La Selva Biological Station in Costa Rica in which the NIMS RD system was used to quantify the micrometeorological edge effects of a primary forest boundary.

Our largest number of NIMS deployments have been in the context of contaminant transport in rivers and streams. For example, we have worked on an approach for the characterization of urban stream quality and algal dynamics where we are developing and integrating methods of urban stream monitoring to provide a good understanding of what biologically relevant, physicochemical stream conditions are causing algal impairment at a given site.  Further discussion of NIMS applications appear in the Contaminant Transport and Aquatic sections.