Optimal LQG Control and Stability of Networked Robot System with Data Dropout

By modeling the networked robot system (NRS) that drops packets randomly, we considered the problem of optimal linear quadratic Gaussian (LQG) control and analyzed the stability of a NRS. We presented a mathematical model based on a packet-based setting, extended the familiar LQG separation principle that allows us to solve this problem using a standard LQR state-feedback design, proposed an optimal algorithm irrespective of the packet drop pattern by constructing an encoder for the unreliable channel and designing the decoder that uses the information it receives across the link to construct an estimate of the state of the networked robot. For the case of packet drops occurring according to a Markov chain, the stability analysis was carried out. Because the separation theorem for linear systems and quadratic cost does not apply to the general framework of NRSs, we used the uncertainty threshold principle to show that under certain conditions there was a rate for dropped packets for which an undisturbed networked control system with imperfect state observation was mean square stable, used a sub-optimal method to simplify the calculation of the estimator and controller, got the solution to the Riccati-like equation and guaranteed the mean square stability of the NRS with perfect state information. This design does not assume any statistical model of the packet drop events and can be implemented as a small modification of an existing LQG control design.