|M.Sc Student||Malachi Liran|
|Subject||Decentralized optimal control of multiple haptic devices|
|Department||Department of Mechanical Engineering||Supervisors||Professor Leonid Mirkin|
|Dr. Maxim Kristalny|
|Full Thesis text|
Teleoperation systems are used to expand operator's abilities to remote environments. In such systems, the operator uses an interface device, “master”, to manipulate a remote robotic system, “slave”. Bilateral teleoperation notion refers to the case in which haptic feedback is provided to the operator in order to create a telepresence experience, i.e., the ability to “feel” the remote site. This work concentrates on a control problem arising in cooperative bilateral teleoperation systems (CBTS), i.e., in the case when two operators (or more) are manipulating two slave devices (or more) in a shared task environment. A situation like this may occur, for example, in a surgery with dual master consoles.
Some of the main challenges associated with such systems arise from communication delays, which in many practical cases may not be neglected. Communication delays may have a destabilizing effect and complicate the analysis by rendering the system decentralized. The most common approach for this kind of problems is based on passivity. This approach guarantees stability regardless the communication delay length. However, the use of passivity methods relies on ad hoc control structures that are restrictive in terms of performance. Moreover, the assumptions required for the use of passivity based methods might not be met in many situations.
The development of alternative optimization-based methods for bilateral teleoperation control is impeded by the decentralized nature of this problem. However, it was shown, recently, that under a set of mild assumptions the optimization problem associated with CBTS is “quadratically invariant”. This observation leads to a novel non-restrictive stabilizing control architecture and serves as a starting point for the current thesis. In this work, the proof of stability for the aforementioned control architecture is completed. Then, Youla parameterization of all stabilizing controllers is used to cast the associated H2 optimization as a decentralized model matching. Inherent structural properties of CBTS are used to split this problem into three independent centralized optimizations, whose solutions are derived using the recent loop-shifting techniques. Due to numerical difficulties, the resulting approach in its general form is suitable only for problems with small delays. It turns out, however, that it is possible to overcome this issue by slightly restricting the design criterion. It is shown that, in most practical cases, the proposed restriction does not limit the performance, but allows to split the original problem into nine tractable optimizations. This results in an intuitive method for performance-oriented controller design, whose potential is demonstrated with several case studies.