Control Without Trust: Risk of Ambiguous Perception ( ONR Tracking:23-000004438)
Abstract
This proposal focuses on the challenge of handling inherent stochastic uncertainty in perception-based decision-making and control for autonomous systems with onboard perception units. The proposal aims to develop a novel multi-stage perception-based control paradigm based on distributionally robust operators, which considers all possible probability distributions consistent with previously observed data and constructs corresponding ambiguity sets to enable effective handling of uncertainties. The proposal presents an integrated modeling framework that incorporates sensing, perception, decision-making, and control as optimization problems, optimized against all probability distributions consistent with prior knowledge, resulting in distributionally robust optimization problems. The proposal addresses fundamental challenges such as decision-dependent ambiguity, uncertainty propagation, and tractable approximation schemes for the proposed models, with the goal of improving the performance and safety of perception-based control systems. By enabling effective handling of ambiguous perception, the proposed methodology has the potential to significantly enhance the robustness and reliability of autonomous systems. The proposed risk-averse perception-based control methodology can minimize the risk of failures, making autonomous systems safer and more efficient.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 11, 2023
- Source ID
- N000142312779
Entities
People
- Nader Motee
Organizations
- Lehigh University
- Office of Naval Research
- United States Navy