Conventional and Explicit MPC Applied to Robotic Systems: a Computational Cost Evaluation

Abstract

The computational cost of implementing Model Predictive Control (MPC) restricts its use in robotic systems with fast dynamics. The use of optimization algorithms to find a control solution is responsible for most of the computational cost. Explicit MPC (eMPC) solutions aim to overcome the computational cost problem by performing the optimization process before runtime. The main contribution of this work is to compare the computational cost of the conventional MPC to the computational cost of eMPC, in the quadrotor stabilization task. The controllers are evaluated in two different scenarios, one in a Model-in-Loop (MIL) environment and the other using the Robot Operating System (ROS) in a Processor-in-Loop (PIL) setup. The results show that eMPC is an alternative to implement predictive controllers with determinism in computational time execution, due to the use of only algebraic operations at runtime, without any numerical tools, but, with the disadvantage of using a significant amount of memory.