C1160: An Integrated Approach to Space Situational Awareness
Abstract
This project has led to fundamental advancements in the fields of Filtering/ Data Assimilation, Multi-Target Tracking and Data based Control. In Filtering, we have developed particle filters that are immune to the Cure of Dimensionality, called the Particle Gaussian Mixture Filters (PGM). We have also developed hybrid consensus and covariance intersection based distributed estimation algorithms that retain the robustness of CI while recovering the performance of consensus methods. In Multi-Target tracking (MTT), we have unified the hitherto thought to be different MTT techniques based on the classical MultiHypothesis tracking (MHT) and Random Finite Set (RFS) based methods. We have developed a highly efficient randomized technique for MTT, called the Randomized Finite Set Statistics (RFISST), that significantly outperforms classical MHT methods based on Munkres/ Murty's algorithms. The efficacy of the techniques has been shown for Space Situational Awareness (SSA) problems. The project has also contributed to the development of a Dynamic Data Driven Approach to Planning and Control of unknown systems/ Reinforcement Learning termed Decoupled data based Control (D2C) that offers a new and highly efficient paradigm for the feedback control synthesis of highly nonlinear and high dimensional systems, problems that were hitherto thought to be intractable. This technique breaks Bellman's infamous "Curse of Dimensionality" in nonlinear feedback control.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 02, 2021
- Accession Number
- AD1137007
Entities
People
- Suman Chakravorty
Organizations
- Texas Engineering Experiment Station