Overview

The introduction of mobility in sensor networks has generated a variety of research topics. We study the problem of sampling and reconstructing the two-dimensional sunlight field under a forest canopy using robotic sensors that can quickly move to designated locations to collect light intensity measurements. The sampling and reconstruction process is carried out in adaptive steps. During each step, the most desirable sampling sites are selected from a pool of site candidates based on a maximum a posteriori (MAP) test. Source statistical models and field roughness are used to further account for heterogeneity.

Approach & accomplishments

The NIMS (Networked InfoMechanical System) platform is depicted in the following Figure. Sensors can move horizontally and vertically to prescribed locations to collect sunlight intensity samples. In our approach, the sampling locations are adaptively selected.

Figure 1. A NIMS platform in the natural environment

Figure 2. The block diagram of the adaptive sampling algorithm

The block diagram of our adaptive algorithm is shown in Figure 2. It consists of the following major components:

• Sampling candidates: are generated from the Delaunay tessellation of the two-dimensional area based on the existing sampling sites. These sampling candidates form a candidate pool, from which new sampling sites are selected.
• Field reconstruction: is conducted using the thin plate spine, which is the preferred method for constructing two-dimensional field from scattered data sites.
• Adaptive sampling site selection: approximates a MAP test to pick data sites that have a high probability of failing the fidelity requirement.
• Statistical model: constructed from extensive field experiments; is incorporated in our adaptive algorithm to aid sampling site selection.

We use the camera to capture snapshots of the sunlight fields. Simulations are conducted on these two-dimensional fields reproduced in simulation. The following figure compares our adaptive algorithm to two other methods: uniform sampling and the Q method. The sunlight field reconstructed using our adaptive algorithm approximates the true field fairly well.

Figure 3. The true and reconstructed sunlight fields.

Future Directions

This project was completed, with the graduation of the graduate student researcher.

People

PI: Greg Pottie

Participants: Huiyu Luo, graduate student