FULL PDF
Abstract
This study presents a comparative analysis of decision-making algorithms employed across autonomous and semi-autonomous system architectures within the fields of transportation, robotics, and industrial automation. We evaluate the structural, computational, and real-time performance dimensions of various algorithms, such as Markov Decision Processes (MDPs), Reinforcement Learning (RL), and Heuristic-based Decision Trees (HDT). By integrating findings from cross-domain applications, we assess algorithmic suitability based on adaptability, interpretability, and risk handling. A mixed-method approach is utilized to synthesize quantitative benchmarks with qualitative operational analyses. The results emphasize that while MDPs show optimality in constrained environments, RL algorithms outperform others in dynamically uncertain contexts. Our analysis also highlights the practical limitations of algorithm portability between domains due to task complexity and safety-critical considerations.
View more >>