CAREER: Collective behavior in multi-agent systems with active sensing

点击翻译按钮获取中文摘要

This Faculty Early Career Development Program (CAREER) project will study how large groups of individual agents may collectively use active sensing to improve on the performance of any one individual. In particular, the project seeks to understand and emulate the way in which swarms of bats use echolocation, which is a type of active sensing. In active sensing, for example sonar or lidar, an agent uses the reflection of an emitted signal to learn about its surroundings. Passive sensing, in contrast, relies on the reflection or emission of signals originating elsewhere, such as vision systems that rely on ambient light, or thermal imaging systems that detect heat. In large bat swarms it has been observed that some members seem to navigate by "eavesdropping" on signals emitted by other bats, instead of generating their own. This observation raises many possibilities, including questions of whether eavesdropping changes the behaviors that the swarm can achieve. This project will address these questions through studies on wild bat swarms, then will employ the results in experiments on multiple mobile robots. Of interest will be the ways that simple rules for active sensing by the collective may give rise to complex emergent behavior. The results of this project will benefit national prosperity and welfare by enabling new capabilities in multi-agent networks of mobile robots, for applications such as search and rescue, surveillance and environmental monitoring, package delivery, and construction. This work will also provide insight into the native behavior of bats, many species of which are critically imperiled due to disease and habitat loss. The project includes engagement with K-12 students across Virginia, and supports the development of educational modules that will be disseminated to classrooms throughout the state. The role of sensing in the collective behavior of multi-agent systems -- and particularly the impact that active and passive sensing may have on emergent phenomena -- is currently unexplored. Robotic systems are increasingly turning to distributed approaches, therefore understanding how to leverage their interaction over sensing channels could represent a paradigm shift in the control of such systems. This project seeks to provide a new vision for the dynamics and control of multi-agent systems by coupling sensing and communication. This project is motivated by the intercepted sensing observed in bat swarms. The research team will collect quantitative data from field experiments with wild gray bats, perform model-free analyses to determine how collective sensing and behavior manifests in bats, use this insight to inform and validate a model of group behavior with active sensing, and create numerical and robotic testbeds for hypothesis-based exploration of these novel collective dynamics.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.