Design and Evaluation of a Machine Learning-Based Decision-Making Framework for Robotics

Abstract

Decision-making systems are of utmost importance for the autonomy of robotics so that robots may negotiate adaptive and complicated environments with very little human interference. This research presents a detailed study of and development of an autonomous robotic decision-making system through generalized AI techniques such as Neural Networks, Support Vector Machines, and Decision Trees. The design of a computer system capable of interpreting a variety of sensory inputs consisting of obstacle distance, battery levels, wheel speed, type of terrain, avenue weather conditions, ambient temperature, and robot tilt constitutes the main objective of this research so that it may dependably decide on the best action to take during a robotic mission. The study uses a carefully chosen dataset with 500 balanced observation points produced from realistic robot operations. Various advanced data preprocessing techniques, such as normalization, noise removal, and feature selection, were performed to improve the quality of the models. Performance-wise, the ANN model proved to be the most accurate (99%), most precise, and had the highest recall, thus clearly outperforming SVMs and DTs. Nevertheless, the SVM model especially finds its use in working with system requirements that need classification, interpretability, and computational efficiency, whereas, in contexts where safety is paramount, a DT model provides clear and transparent decision logic. The study advances the paradigm of AI-driven decision-making to solve intermediate-level decision problems for autonomous, flexible, and efficient robotic operations. More lines of research could go into hybrid AI and real-time implementation in critical sectors like disaster response and hazardous material management.