Hot-start efficiency of quadratic programming algorithms for fast model predictive control: A comparison via an adaptive optics case study

Model Predictive Control (MPC) with fast sampling rates can be extremely demanding in terms of the required computational time. However, the control problem in some cases does not change much from one sampling instance to the next, and therefore hot-start can be used to considerably accelerate the solution of an online optimisation problem. An adaptive optics system is used in this work as an example of such a control system to evaluate the benefits of hot-start for different families of optimisation algorithms. A comparison of the computational times and a discussion of hot-start efficiency for Interior Point, Active Set, Gradient-based and Augmented Lagrangian algorithms are provided in this contribution.