Rapid Control Selection through Hill-Climbing Methods

TL;DR

This paper explores the use of hill-climbing algorithms as an efficient alternative to traditional search methods for control selection in complex dynamical systems, demonstrating improved search efficiency especially in high-dimensional control spaces.

Contribution

It introduces and evaluates a hill-climbing approach for control selection, showing its advantages over systematic and random search methods in complex scenarios.

Findings

Hill climbing improves search efficiency in high-dimensional control spaces.

Search times remain stable despite increasing control configurations.

Hill climbing outperforms traditional search algorithms in tested scenarios.

Abstract

Consider the problem of control selection in complex dynamical and environmental scenarios where model predictive control (MPC) proves particularly effective. As the performance of MPC is highly dependent on the efficiency of its incorporated search algorithm, this work examined hill climbing as an alternative to traditional systematic or random search algorithms. The relative performance of a candidate hill climbing algorithm was compared to representative systematic and random algorithms in a set of systematic tests and in a real-world control scenario. These tests indicated that hill climbing can provide significantly improved search efficiency when the control space has a large number of dimensions or divisions along each dimension. Furthermore, this demonstrated that there was little increase in search times associated with a significant increase in the number of control configurations considered.