Fault-Tolerant MPC Control for Trajectory Tracking
This addresses fault tolerance in MPC for trajectory tracking, which is an incremental improvement in control systems for applications like robotics or autonomous vehicles.
The paper tackles the problem of model predictive control (MPC) performance degradation due to faults by developing a strategy that combines active fault identification with trajectory tracking, using methods like Constrained Convex Generators and SVD decomposition for runtime model adaptation.
An MPC controller uses a model of the dynamical system to plan an optimal control strategy for a finite horizon, which makes its performance intrinsically tied to the quality of the model. When faults occur, the compromised model will degrade the performance of the MPC with this impact being dependent on the designed cost function. In this paper, we aim to devise a strategy that combines active fault identification while driving the system towards the desired trajectory. The explored approaches make use of an exact formulation of the problem in terms of set-based propagation resorting to Constrained Convex Generators (CCGs) and a suboptimal version that resorts to the SVD decomposition to achieve the active fault isolation in order to adapt the model in runtime.