A Safe Motion Control Method for Robots Based on Online Interpretable Decision Trees

As one of the most interpretable machine-learning algorithms, decision trees have drawn increasing interest in intelligent robot control in recent years. As robots increasingly operate in dynamic and unstructured environments, the limitations of traditional black-box control strategies in interpretability, safety verification, and online adaptation have become increasingly prominent. Decision trees, with their inherently transparent reasoning structures and strong potential for formal verification, have emerged as ideal candidates for constructing trustworthy robotic controllers. This paper systematically reviews research on decision trees in robot control. Representative works are surveyed across five key dimensions: classical decision tree algorithms, online incremental decision trees, safety-constrained decision trees, interpretable reinforcement learning, and policy extraction. Furthermore, it analyzes the advantages and limitations of existing methods in robot motion control and outlines future research trends in this field.