Socially Networked Autonomy: How Should Machines Interact With Society?

点击翻译按钮获取中文摘要

This project studies decision design methodologies for autonomous agents that are networked with and interacting among human beings in societal systems. The running application example is that of an urban highway network in which a portion of the vehicles are autonomously operated. The core question is this: how should a system planner design the routing policies of the autonomous vehicles to have the greatest positive impact on overall network traffic congestion, even if human drivers react in a self-interested way to the behavior of the autonomous vehicles? Intuitively, it would appear beneficial to design the autonomous vehicles to be altruistic; that is, to select highway routes that cause the least congestion to other drivers. However, recent work shows that this simple and intuitive routing design policy may backfire, leading to a cascade of behaviors which inadvertently worsens congestion. This project will investigate how a designer can circumvent these pathologies through the design of smart routing policies. One of the key questions this project will address is how selfish should autonomous vehicles be? This project includes public outreach to help members of the public discern between good and bad modes of influence in socially networked autonomous systems. In addition, it will provide a theoretical framework for engineering practitioners to certify that the interactive aspects of smart systems have been designed to provide broad societal benefits in a principled way.This project addresses fundamental behavior design questions for socially-networked multiagent systems; an intrinsic feature of this problem is that overall system performance is governed not only by the designable behavior of the autonomous agents, but also on the self-interested behavior of human participants. The element of interactive optimization places this project squarely in the domain of game theory, and the investigators adapt popular congestion game models to study these questions in the context of smart autonomous vehicle routing in highway networks. A fleet of autonomous vehicles is modeled as a discrete player in a continuous-player-set network congestion game; the project studies how routing policies for the autonomous vehicles impact the game-theoretic equilibria of the overall system. This allows the investigators to model, in a principled way, the interactions between designed agents (i.e., autonomous vehicles) and selfish agents (i.e., human drivers). Ultimately, the investigators seek a comprehensive understanding of how and when autonomous vehicles should be endowed with various routing policies. This project advances knowledge in two distinct dimensions: first, it provides an in-depth case study examining the opportunities and limitations of the use of connected autonomous vehicles for influencing traffic routing in transportation networks. Second, and more broadly, it will provide the foundations for an underlying theory of social influence in smart and connected cyber-physical systems. As smart cyber-physical devices are increasingly integrated into public life, it is critical that these smart devices are designed to interact with and influence human behavior in ways that are effective, efficient, and principled – and this project represents a central pillar of this endeavor.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.