PARALLEL NONLINEAR STOCHASTIC CONTROLS USING PROBABILISTIC REPRESENTATION OF PARTIAL DIFFERENTIAL EQUATIONS
Abstract
In this research, stochastic control and machine learning algorithms are proposed for single and multi-vehicle autonomy in sea environments. The research includes the development of novel planning algorithms using sampling-based nonlinear stochastic control. These algorithms will relyon the connections between backward partial differential equations and expectations evaluated on nonlinear stochastic processes. Implementation of the algorithms will be performed on GPUs to achieve real time execution. In addition to decision-making algorithms, machine learning methods for learning dynamics are also proposed. These methods are based on Feed Forward and Recurrent Neural Network representations, which will be used to learn the underlying dynamics of vehicles operating in sea. An important capability for autonomous vehicle operating in naval environments is uncertainty quantification in decision making and learning. Furthermore, state estimationalgorithms will be used for sensor integration and distributed localization.The research will be executed by one graduate student at Georgia Tech in collaboration with the Naval Surface Warfare Center, Panama City Division. Algorithmic and theoretical work will be primarily performed at Georgia Tech while testing and evaluation will be performed in the testing facilities at Panama City Warfare Center.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 15, 2021
- Source ID
- N000142112074
Entities
People
- Evangelos A. Theodorou
Organizations
- Georgia Tech Research Corporation
- Office of Naval Research
- United States Navy