We address the problem of monitoring spatiotemporal phenomena at high fidelity in an unknown, unstructured, dynamic environment. Our thesis is that a cooperative system comprised of mobile robots (suitable on their own for efficient temporal sampling) and a static sensor network (suitable on its own for efficient spatial sampling) is an effective means of addressing this problem. We provide theoretical as well as empirical evidence supporting this thesis by decomposing the design of the proposed collaborative system into three constituents. These are respectively, robot coverage and exploration through sensor network deployment, sensor network-assisted robot navigation and sensor network-mediated multi-robot task allocation.
The first subproblem we address is the embedding of an active infrastructure (sensing, communication and computation) into the environment using robots, which simultaneously use this infrastructure for coverage and exploration. Our algorithm for this is provably complete, decentralized, scalable, robust, fault tolerant and can be used on simple robots. We present experimental and simulation results which verify the performance of the algorithm. We also present theoretical results which illustrate its asymptotic behavior.
Once a network is deployed it can be used for robot navigation. We present an algorithm that allows robots to navigate precisely and reliably using a deployed sensor network. This networkdirected navigation approach can be used by simple (modest computation, communication and sensing requirements) heterogeneous robots. Extensive empirical testing conrms the validity of our approach.
The final subproblem we address for efficient spatiotemporal monitoring is multi-robot task allocation. The network is used as a spatially diverse 'sensor' for event detection and as a mediator which assigns and navigates robots to high-value sampling locations. This builds on the deployment, coverage, and navigation capabilities presented earlier. Network-mediated task allocation allows robots to 1. respond to task they cannot directly sense, 2. communicate at long ranges, 3. efficiently repair and maintain network. We validate our algorithm for task allocation in field and lab experimental settings.