Abstract

This work reports the design and implementation of Model Predictive Controllers using both implicit and explicit formulations that leverage FPGA acceleration to reduce the execution time of computationally demanding optimization tasks. A case study of a servomotor is used as a driving example to illustrate the development process for each formulation. The implementations of complete control loops are experimentally validated using a dSPACE rapid prototyping platform. Experimental results show that, by leveraging FPGA acceleration for demanding optimization tasks, the controller using an implicit formulation for the servomotor achieves a control interval of less than 20 µs. Moreover, an explicit formulation using artificial neural networks for approximating the control law futher reduces the latency of the control loop to sub-microsecond levels. Overall, the results highlight the effectiveness of FPGA-based acceleration for implementing low-latency, real-time Model Predictive Controllers and the potential of these techniques for addresing problems with stringent timing constraints. © 2025 IEEE.