Model Predictive Control and Communications Resource Assignment
in Coordinated Systems
Prof. Robert Bitmead (University of California, San Diego)
An approach to the control of coordinated autonomous systems will be
presented in which interactions between members will be managed by
specifying constraints. The key points of the approach rest on the
following ideas:
Global coordination tasks split into reference trajectory
specifications for each member and communications resource assignments
between members.
Local tasks are those which are real-time and include determining local control laws to track the reference trajectory and computing estimates of neighbours' states from communicated information.
Interaction between systems is managed by specifying constraints, such as "no collision," which yield a natural design paradigm.
Model Predictive Control provides a local control law approach amenable to handling constraints and is equipped with nascent guarantees of stability, feasibility and robustness.
Imprecision in the position or state of a neighbour may be accommodated by tightening the interaction constraints in accordance with the degree of uncertainty.
Communication resource assignment (bit-rate and message frequency)
between the interacting systems determines the uncertainty of the
state of the neighbors.
These ideas will be presented and, for a vehicle coordination problem, shown to yield a simple convex optimization problem based on state estimation, whose solution (if it exists) provides; a feasible assignment of bit-rates to the signals on the links, estimator feedback gains, and steady-state covariances for the neighbors' state estimates. This ties together the global system control objective with the communications assignment in a tractable, scalable fashion