Stochastic LQG Algorithms For Robot Dynamic Tracking Over Wireless Network

Compared with conventional point-to-point control systems, Network Control Systems (NCSs) is completely distributed, thus it can offer many advantages such as realizing remote operation and control, sharing resource, easily installation and maintenance, a high diagnostic capacity, increasing the system flexibility and reliability, etc.. The existing control networks are usually based on wired transmission medium, so a series of problems such as cable deployment on surface and wall, and the network maintenance. The fixed cables take much trouble to the applications when some equipments needs to move. By using wireless communication technology into network control systems, Wireless Network Control Systems (WNCSs) allows fully mobile operation, flexible installation and rapid deployment, while reducing maintenance costs. However, building a wireless networked control system is a challenging task because a wireless network inevitably introduces random delays and packet loss in the feedback loop.In this paper, firstly, the definition of WNCSs and essential problems are described in this dissertation, and the research progress of WNCSs are also introduced from the perspectives of control theory and communication theory; then, the dynamic tracking problem of a wheeled mobile robot is presented, and the mathematic description is given; the necessity of a new algorithm to deal with the varying time delay and packet dropout is shown through a simulation about wireless network's characteristic.Secondly, while taking the delays and packet dropout into account, a continuous linear time-invariant systems'networked model is established. Then an algorithm that combines a stochastic optimal control scheme with predictive control is proposed. The algorithm consists of two parts:the stochastic optimal control is used to deal with the delay which less than a sampling period, while the predictive control to give certain degree of compensation for delay which is more than a sampling period or the packet dropout. Finally, this algorithm is designed as a tracking controller. The simulation shows that this algorithm has a good performance.Then, two control loops that share a communal wireless network are considered. Each loop contains a robot dynamic tracking problem, then a bandwidth schedule modular is designed based on the feedback theory. From the True Time simulation results, the control loop with a bandwidth schedule has a good performance and the control loop without bandwidth schedule may be not able to realize dynamic tracking.Finally, the results in this dissertation are summarized.