There are a myriad of current applications in which robotic sensor networks are having an enormous impact. Examples include a network of coordinated underwater gliders tracking the motion of chemical pollutants, a camera network monitoring a busy street, or a group of autonomous vehicles providing force protection. At the most basic level, coordination strategies for these networks involve agents repeatedly taking measurements, communicating with other agents, processing the collected data, and taking actions in response. A common assumption throughout these applications is the continuous or periodic availability of information to the agents about the state of other agents and the environment, and the synchronous execution of these strategies. This synchronization assumption poses nontrivial challenges in practice and leads to inefficient implementations in terms of processor usage, communication bandwidth, and energy. Periodic communication, for instance, may lead to a wasteful use of the available resources. When asynchronism is considered, it is often done in "a posteriori" fashion: guarantees only hold if the agents time schedules satisfy conditions ensuring the freshness of information. While such results are valid from an analysis viewpoint, they are unsatisfactory from a design perspective because ensuring that such conditions hold is not built into the algorithm synthesis.
The objective of this proposal is the design of self-triggered coordination strategies that account for uncertainty in the state of other agents and the environment, and are able to produce substantial energy savings in the network operation. The key conceptual novelty is the study of how the performance of the overall network task is affected by the quality of the information available to the agents. This understanding leads to tools and triggering criteria for individual agents that allow them to autonomously decide when they need fresh information to successfully perform the required task. Self-triggered strategies eliminate the need for continuous communication, sensing, and re-planning, incorporate uncertainty at the control design stage, seamlessly handle asynchronous executions of plans, and increase agent autonomy and network efficiency.
Intellectual Merit: This proposal seeks to develop tools, abstractions, and techniques that help design self-triggered cooperative strategies for autonomous robotic sensor networks. Our ultimate goal is to synthesize robust and efficient cooperative strategies that handle uncertainty and asynchronism, and ensure that the robotic network performs the assigned task with guaranteed quality of service, while operating with limited energy supplies, bandwidth, and computational resources. The research plan is structured along the following thrusts: (i) the synthesis of reliable models and abstractions that capture the uncertainty about the state of the network and the environment; (ii) the identification of triggering criteria that allow agents to determine the impact that the actions planned with their current information have on the performance of the network; (iii) the development of stability and correctness tools suited for the analysis of self-triggered coordination strategies and the precise characterization of their robustness and efficiency properties. We envision that the proposed paradigm will lead to the synthesis in a variety of distributed scenarios of novel coordination algorithms that optimize the trade-offs between performance and implementation cost and have superior robustness guarantees than existing strategies.
Broader Impacts: Multi-agent systems are extending the range of human capabilities in an increasing number of scenarios, including the study of oceans, disaster recovery, environmental monitoring, and surveillance. The results of this project will help design robust and efficient cooperative strategies that naturally account for uncertainty and are able to produce substantial energy savings in the network operation. The educational activities are integrated into the research plan and consist of (i) undergraduate student involvement in research via summer internships, independent study courses, and senior-design projects. The PI belongs to the UCSD Cymer Center for Control Systems and Dynamics and will be involved in the supervision of industry-sponsored undergraduate research in control; (ii) offering of a graduate course on cooperative control and graduate student supervision; (iii) outreach targeted at high school students and teachers through the California State Summer School for Mathematics and Science program and collaboration with the NSFfunded project ComPASS at UCSD; (iv) involvement of minority students through participation in the activities of the UCSD School of Engineering IDEA Student Center; (v) broad dissemination activities (journal publications, conference, workshop presentations, and conference organization).
