CAREER: Optimal Transport-based Density-Aware Multi-Agent Exploration

点击翻译按钮获取中文摘要

Multi-robotic systems can provide many benefits over single robots in exploring and servicing large-scale environments, such as in problems of search and rescue, surveillance and reconnaissance, smart farming, infrastructure inspection, wildlife monitoring, weather monitoring, and planetary exploration. However, the deployment of multiple robots in an efficient manner remains a challenge. Traditional approaches for deploying robots that uniformly or randomly cover a given domain are not necessarily efficient. This Faculty Early Career Development Program (CAREER) award will support fundamental research to develop a new method for multi-agent control by incorporating density information that reflects the priority or importance of covering specific areas in the domain. This paradigm shift from traditional to adaptive coverage creates an opportunity to maximize the efficiency of multi-agent explorations in various missions where both time and resources are limited. The outcomes of this project will include hands-on demonstrations of multi-agent control for wildlife monitoring. The integrated research and educational activities will directly impact upcoming generations of scientists and engineers through involving graduate and undergraduate students in research and organizing education and outreach programs for K-12 students with an emphasis on underrepresented minority groups.This project focuses on improving the efficiency of multi-agent explorations by employing optimal transport (OT) theory as a tool to synthesize multi-agent trajectories. OT provides a way to quantify the distance between two probability density functions (PDFs): the reference PDF that is chosen to indicate the relative importance or priority of the given domain and the current PDF that will be formed based on the time-averaged behavior of the multi-agent system. A decentralized optimal control law for multi-agent systems will be developed to manipulate and reshape the PDF of the multi-agent trajectories to make it as close to the reference PDF as possible. This new framework, based on concepts from OT theory, will serve to advance knowledge of multi-agent control for broad environment exploration.This project is jointly funded by the CMMI-DCSD program and the Established Program to Stimulate Competitive Research (EPSCoR).This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.