Event Driven Online Optimal Control of Autonomous Systems for Mine Countermeasure (MCM) Missions
Abstract
Autonomous underwater vehicles (AUVs) are valuable assets for mine counter measure (MCM) missions. AUVs are tasked with operations in increasingly complex and potentially adversarial environments. Even in uncontested environments with stationary mines, AUVs are challenged by limited situational awareness, and slow and intermittent communication capabilities. These problems are exacerbated for MCM missions involving mobile underwater mines (MUMs) that require target tracking and the need for potential complex target intercept strategies that could involve discrete event-based decisions, while exploiting opportunistic sensing and communication. MUM tracking can also be complicated by natural obstacles (e.g., in harbors and rivers) or by the motion of the MUM and the limited field of view of the tracking sensor, leading to target occlusion and intermittent sensing. The underwater environment also results in uncertain interaction forces/dynamics for the AUV. Moreover, the interaction dynamics between an AUV and a MUM (i.e., how will the MUM react to an approaching AUV) are uncertain, further complicating the intercept strategies. Despite these challenges, limited power and time critical mission aspects motivate the need for optimal guidance and control methods. The broad research objective of this project is to develop online (i.e., real-time) adaptive optimal controllers for maritime systems that operate in complex environments where intermittent sensing and actuation render modern online optimal control methods unsuitable. Efforts leverage the success of our recently developed actor-critic-identifier, model-based approximate dynamic programming (ADP) methods for continuous systems that approximate the optimal control through reinforcement learning. This project generalizes these previous efforts by examining hybrid dynamic systems. Such a generalization is required to enable more practical MCM missions that involve a mix of continuous dynamics (i.e., the dynamics of the physical vehicle and mines) and discrete dynamics (e.g., resulting from intermittent communication events, intermittent sensing events, discrete changes in the mine intercept strategy, etc.). The specific aims that motivate the proposed tasks include: characterization of the stability and the robustness of optimal controllers under intermittent sensing and actuation, development of computational methods for the synthesis of feedback controllers under intermittent sensing and control, and development of efficient methods to enable real-time approximation of optimal solutions to multi-agent mine engagement problems under abrupt network topology changes and communication rate constraints. We propose to achieve the aforementioned objectives via three integrated research tasks. The first task focuses on the development of online approximate optimal control methods for switched and hybrid systems, the second task is devoted to the development of online optimal control methods that rely on model-based extrapolation of state trajectories, in conjunction with intermittently available high-fidelity state estimates, and the third task explores online policy synthesis techniques for countering MUMs using differential game and augmented state formulations of multi-agent herding problems. Outcomes from each task will be validated through rigorous nonlinear analysis methods (i.e., Lyapunov-based methods), and the performance will be demonstrated through simulations. Experimental demonstration of the developed methods will also be pursued in lieu of additional simulation studies, if government MCM equipment is furnished.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 09, 2021
- Source ID
- N000142112481
Entities
People
- Warren E Dixon
Organizations
- Office of Naval Research
- United States Navy
- University of Florida