Online Optimization-Based Feedback Controllers for Dynamical Systems in Stochastic Environments with Partially Known Performance Metrics and Safety Constraints
Abstract
The current trend in the armed forces towards the deployment of large-scale vehicle systems in dynamic, adversarial, and network environments requires the development of new methods and techniques that address the dramatic increase in complexity. This transformation makes it apparent that existing tools lack the computational and communication flexibility to handle information-rich, evolving environments and to enable autonomy. Significant research efforts are needed to enable breakthrough approaches for decision making and learning on timescales matching the dynamics of these autonomous systems. This project develops new approaches for the synthesis of feedback controllers that optimize the operation of autonomous systems while ensuring safety and guaranteeing varying degrees of performance and operational efficiency. The proposed control framework dynamically regulates inputs and states to optimal solutions of time-varying optimization problems that encode operational costs and safety objectives, shifts in operational constraints, and timeÂvarying topologies of the underlying network system. Our technical approach blends techniques from dynamics, control, optimization, and learning in the synthesis of safe optimization-based controllers inspired by first-order optimization methods and equipped with safety filters that leverage control barrier functions. We will explore methods to learn costs, constraints, and physical models from data, and assess the robustness of the proposed framework to learning errors leveraging input-to-state stability and safety certificates. A successful outcome of the research will lead to new modeling and control synthesis approaches for autonomous systems operating in stochastic and adversarial environments. The combination of performance and operational efficiency developed here would benefit many applications of interest to the Air Force, including multi-agent networks, unmanned aerial vehicles, and tactical and information networks.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 14, 2024
- Source ID
- FA95502310740
Entities
People
- Jorge Cortés
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of California, San Diego