Controlling highly dynamic systems, such as legged robots, can be a very challenging task and requires a lot of computational resources. Most modern control architectures use state-of-the-art control strategies involving Model Predictive Control (MPC). These controllers require an underlying Quadratic Program (QP) to be solved in real time using numerical algorithms. Programs implementing those algorithms are called QP-Solvers.
This research focuses on comparing different QP-Solvers on different hardware architectures in order to evaluate the efficiency of each solver regarding speed and CPU power consumption. Combining this information into a “performance per watt” metric makes it possible to compare QP-Solvers across different hardware architectures easily.
When a dynamic system needs to be controlled autonomously, computational and electrical power is limited by the specifications of its onboard computer. Having knowledge about the efficiency of the used QP-Solver can therefore enable tasks, which are more complex or required longer endurance.