The overarching goal of this project is to make multi-robot systems operate reliably in real-world conditions. Most current multi-robot systems are "brittle" since the failure of a single robot may bring the whole team down. Cyber-attacks that incapacitate robots or compromise their sensors are becoming increasingly realistic. This project will focus on multi-robot coordination in scenarios where the robots operate in failure-prone or adversarial environments. The key question that the project will address is how can a team of robots carefully coordinate and adapt their actions to make them resilient to failures. The outcomes of this project will be a significant step towards the overarching vision of persistent and reliable autonomous operations in real-world conditions. These research activities will go hand-in-hand with educational and outreach activities that will integrate real-world examples of multi-robot coordination in precision agriculture and environmental monitoring to train the next generation of engineers in multi-disciplinary thinking. To broaden participation from underrepresented minorities, the project will develop a new hands-on robot programming summer camp for middle-school students. To ensure broad applicability, middle-school teachers will be trained in this workshop with the goal of them incorporating the technical activities in their classrooms. The investigator will develop a new course on multi-robot coordination and mentor undergraduate/graduate students, who are the future faculty, to create and lead a new design-oriented course through the Student Initiated Courses program. Dissemination of our findings will be achieved through workshops, articles geared toward broader audiences, and community outreach events.
The main intellectual contributions of this project are in the introduction of a novel class of problems and development of new algorithms and theoretical limits for multi-robot coordination in adversarial and uncertain settings. The underlying framework will be based on submodular optimization which is an often-used technique in multi-robot coordination. Existing works that use submodular optimization for coordination typically assume that the function value can be computed exactly. The key contribution of this project will be to relax this assumption which leads to a wide variety of research problems that will be investigated. Broadly speaking, three classes of problems will be investigated: devising offline coordinated deployments that are secure against adversarial attacks; devising online, adaptive strategies that are resilient to real-time failures and attacks; and devising online and offline plans that take into account novel measures of risk in stochastic submodular optimization. Various versions of the problem such as: distributed and scalable; richer attack models particularly for distributed networks; and richer utility functions particularly those using deep neural networks, will be investigated. The project seeks to devise algorithms with constant factor approximation and constant competitive ratio guarantees.
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.
