A Framework for Information Theoretic Cooperative Sensing and Predictive Control
Abstract
The research has dealt with decentralized predictive sensing and control in the presence of uncertainties and constraints. In the first part, we have adopted an information-theoretic framework based on general recursive Bayesian estimation. Each agent continuously performs Bayesian updates of the local belief while selecting the control action minimizing an information theoretic cost. Computational intractability of existing schemes with large number of agents has been addressed by partitioning the search space and proposing corresponding tailored optimization algorithms. The second part of the research has focused on constraint satisfaction. We have studied the problem of decentralized control of a network of integrators subject to state and input linear constraints and affected by additive, set-bounded disturbances. We have introduced the notion of decentralized robust control invariant (DRCI) sets and provided a parametrization of such sets in bounds on states and control inputs. We have shown that the set of parameters leading to non-empty DRCI sets is polyhedral, and thus decentralized, constrained robust control design is a convex optimization problem. We have also addressed the problem of averaging the state of each network element and proposed an asymptotically stabilizing algorithm which is non-iterative and does not require centralized design procedure.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 11, 2012
- Accession Number
- ADA577143
Entities
People
- Claus Danielsson
- Francesco Borrelli
- Karl J. Hedrick
- Mark Godwin
- Miroslav Baric
Organizations
- University of California, Berkeley