UC Berkeley

In light of the current surge in mobile robotics development, receding horizon control has become of high interest to both academia and industry. This control technique is a systematic way to compute intelligent control actions while accounting for complicated system dynamics, forecasted environment information, and system constraints. This work contributes to the field of receding horizon control by merging sliding surfaces into the controller design process. The resulting methodology is named receding horizon sliding control. It is shown that the invariance properties of sliding surfaces can be exploited for deriving provably stable and persistently feasible controllers for a wide class of constrained nonlinear systems. Furthermore, for linear systems this work reduces the complexity of receding horizon tracking controllers compared to current state-of-the-art methods by exploiting the flatness of sliding hyperplanes. The practicality of the proposed control approach is demonstrated by using applications from vehicle dynamics. The applications include an autonomous underwater robotics problem, an automotive engine control scenario, and a self-driving vehicle control case study. Simulations and real-world experiments confirm the effectiveness of the developed control methodology. The results indicate that the proposed control scheme is typically easy to tune, behaves well under system uncertainty, and has manageable computational requirements that make it amenable to fast systems with sampling times of the order of fractions of seconds.