Active set updates in event-triggered networked MPC

We show how to exploit the piecewise affine structure of the optimal control law in networked linear-quadratic MPC. The complete explicit solution is not required in this approach, but the current piece, i.e., the feedback law and its polytope of validity, is computed on demand instead. Assuming such a piece has been determined for the current state, a lean local computational node suffices to check whether future states are still in the current polytope and to evaluate the optimal affine feedback law if that is the case. Whenever the system leaves the current polytope, the local node requests a central, powerful node to determine a new control law and polytope by solving a MPC problem, and to send the resulting law and polytope over the network. In this contribution we extend the approach by not sending the information for a single control law, but for multiple control laws for the predicted system behavior. We show that the number of requests and the amount of transmitted data can be reduced by using active set updates on the local node.