Summaries - Office of Research & Innovation
Research Summaries
Back Robust Modeling and Control for Adversarial Autonomy: Data Driven Approaches
Fiscal Year | 2021 |
Division | Graduate School of Engineering & Applied Science |
Department | Mechanical & Aerospace Engineering |
Investigator(s) |
Kaminer, Isaac I.
Clark, Abram H. |
Sponsor | Office of Naval Research (Navy) |
Summary | The goal of the proposed work is to develop a new mathematical and computational framework for agent-based modeling and control in adversarial autonomy scenarios using data-driven methods. These agent-based simulations mimic real, physical autonomous agents, where the movement of each agent is determined (at least in part) by the motion of the other agents. However, in adversarial environments where agents may be destroyed, the survival of each agent is inherently \textit{probabilistic}. The key idea of the proposed work involves merging deterministic agent-based simulations with data-driven, probabilistic performance metrics. This is non-trivial to do, since the survival or destruction of any agent will have direct consequences for all other agents in the scenario. We will leverage previous work, where we have used large-scale agent-based optimization to determine optimal defense strategies given an observed attacking swarm, including simple heuristic models that couple the deterministic and probabilistic elements of the optimization. The proposed FY21 research will leverage these tools to create a new {\bf hybrid modeling and optimization framework} where adversarial autonomy scenarios can be faithfully modeled and optimal control strategies can be formulated. In particular, we will show that the proposed framework deals with a large class of optimization problems that involve data-driven performance metrics/optimization cost functions and physics-based dynamics of the agents involved. This will include swarm-on-swarm engagements as well as other autonomous missions performed in adversarial environments. As an example, we propose to apply this framework to analyze performance of the cooperative surveillance mission by multiple agents in the presence of threats. By leveraging our previous work on threat modeling in a swarm context, we will show that the proposed framework is flexible enough to handle mobile and stationary threats in the cooperative surveillance problem. Our framework can also be used to study whether redundancy or adding defenders improves mission performance. Additionally, in line with our previous work, we will continue to apply this hybrid modeling and optimization framework toward universal counter UxS strategies for adversarial engagement which are effective against a wide variety of tactics and scalable to large-scale swarms. These strategies will provide guarantees against black box adversaries with unknown internal control algorithms. Finally, we propose to address the "scaling break point'' problem, or the point at which failure modes emerge in swarms as they become very large, in the context of adversarial swarms. For example, when does increasing the numbers of the attackers and/or defenders becomes counter-productive? Are there laws of large numbers for adversarial swarm engagements? We plan to exploit the techniques developed in FY20 as well as the formalism discussed above to answer these crucial questions. |
Keywords | Swarm UAV Swarm vs Swarm Autonomous systems |
Publications | Publications, theses (not shown) and data repositories will be added to the portal record when information is available in FAIRS and brought back to the portal |
Data | Publications, theses (not shown) and data repositories will be added to the portal record when information is available in FAIRS and brought back to the portal |