Inherent Trade-offs in Multiagent Coordination
Abstract
Inherent Trade-o s in Multiagent Coordination Project Summary Jason R. Marden Department of Electrical, Computer, and Energy Engineering University of Colorado at Boulder Many systems can be characterized by a collection of interacting subsystems each making local decisions in response to local information. One example is the problem of reconnaissance and surveillance where the goal is to allocate sensors to a given mission space to maximize the probability of detecting a catastrophic event. Regardless of the speci c application domain, the primary goal in such systems is to design local control policies for the individual subsystems to ensure that the emergent collective behavior is desirable with respect to the system level objective. Distributed decision-making architectures provide the foundation for designing systems with robust and vast capabilities that are of utmost importance to the Department of Defense; however, it is important to highlight that such architectures also impose constraints on achievable performance guarantees. Here, performance guarantees of interest in multiagent systems span a wide variety of measures including the e ciency of stable solutions, rates of convergence, resilience to adversarial intervention, among others. The proposed work will focus on the following three directions aimed at characterizing the inherent trade-o s in multiagent systems with regards to these performance measures. The Impact of Information in Multiagent Systems: The information available to the individual agents, either attained through communication or sensing, invariably de nes the space of admissible control laws. Hence, informational restrictions impose constraints on the achievable performance guarantees in multiagent systems. Providing agents with additional information clearly could lead to new control designs with improvements in the e ciency of the emergent collective behavior; however, it turns out that such gains frequently come at the expense of the underlying convergence rates. The rst technical direction will seek to formally characterize the apparent trade-o between information, e ciency, and convergence rates in multiagent systems. Resilience to Adversarial Interventions: Transitioning from a centralized decision- making architecture to a distributed decision-making architecture provides a door through which such systems can be manipulated by adversarial interventions. The second technical direction will seek to characterize how the structure of the decision-making architecture, e.g., enmeshment of the agents control laws and objectives, impacts the resilience of a multiagent system to such interventions. Characterizing the Potential Impact of Leadership: Many multiagent systems of interest to the DoD involve the integration of both human and computer controlled decision- making entities. The motivation for this integration centers on the potential of human decision-making entities to improve the operational e ciency of a given multiagent system by acting as a coordinating entity. The third technical direction of this proposal will seek to characterize the potential impact of human-intervention in multiagent coordination with regards to the e ciency of the emergent collective behavior.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2016
- Source ID
- N000141512762
Entities
People
- Jason R. Marden
Organizations
- Office of Naval Research
- Regents of the University of Colorado
- United States Navy