Decentralized Optimal Control of Cooperating Networked Multi-agent Systems
Sponsor: National Science Foundation
Award Number: ECCS-1931600
PI: Sean Andersson
Co-I/Co-PI: Christos Cassandras
Abstract:Multi-agent systems encompass a broad spectrum of applications, ranging from connected autonomous vehicles and the emerging internet of cars, where the spatial domain may be hundreds of miles with time horizons over hours of days, to micro-air vehicles which operate over meter length and minute time scales, and down to nano-manipulation with nanometer spatial microsecond time resolution. This project seeks to address five key challenges in networked multi-agent systems: (1) scalability, necessitated by the increasing large-scale nature of the networked systems being designed, (2) autonomy at the individual level, required to ensure a resilient and secure system, (3) communication that is secure and efficient, particularly crucial in wireless settings where the agents have limited energy resources, (4) avoiding local optima that arise from the complex nature of the system interactions and which may yield poor performance, and (5) exploiting real-time data, taking advantage of the modern reality of data-rich environments. While the core of the project is centered on a theoretical approach that extends over the diverse length and time scales needed, it also includes experimental validation using robotic platforms that will provide a platform to showcase and communicate results to a broad audience.
The scope of the proposed project is captured through a general optimization (both static and dynamic) framework which encompasses the vast majority of interesting problems faced by researchers and practitioners. Within this framework, we will pursue three specific tasks: (1) Develop on-line solutions for dynamic optimization problems in networked multi-agent systems, (2) Determine when decentralization without sacrificing the performance of a centralized solution is possible and develop explicit decentralized control algorithms even in cases where some performance degradation is needed, and (3) Address the challenge of multiple local minima in the optimization through the use of boosting functions to escape those local optima. The intellectual merit of these tasks lies in three conceptual cornerstones: (1) Replacing the traditional time- driven paradigm with an event-driven approach, allowing for algorithms whose complexity grows with the number of such events, not the state dimensionality of the network, (2) Using a data-driven approach to optimization, allowing for an approach which can handle the increasing complexity of real-world systems where traditional approaches based on elegant but often inadequate classical models fail, and (3) Escaping local optima in distributed optimization, where the use of novel mechanisms for escaping those local solutions overcomes the limitations inherent to gradient-based approaches. The project is built upon a framework for networked multi-agent systems that is extremely broad, encompassing sub-problems such as coverage control, consensus, persistent monitoring, and optimal formation control, and application domains from connected automated vehicles down to nano-manipulation. As such, our research will advance the state-of-the-art in all domains that rely on networked systems. In addition, specific tasks on education and outreach will be pursued, including hosting rising high school seniors in the labs of the PIs for a summer research internship, showcasing the results to middle and high-school students through demonstrations with mobile robots, and engaging undergraduate students in research.
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.
For more information: click here.