Performance Comparison of Blocking vs. Non-Blocking State Machines on a Semi-Humanoid Robot

Abstract

State machines serve as the core modeling paradigm in embedded and robotic control systems. The architectural selection between blocking and non-blocking implementations under multithreaded control environments critically determines system responsiveness and development complexity, yet existing studies predominantly rely on qualitative theoretical analyses without quantitative evidence derived from physical experimental platforms. This study employs a competition-grade semi-humanoid robot as the experimental vehicle and constructs a customized test surface featuring continuous grayscale gradients with randomly placed obstacles to emulate complex terrain conditions. Two distinct patrol state machine programs, one blocking and one non-blocking, are independently designed and implemented on this platform. An offline frame-by-frame analysis methodology is adopted to precisely extract the frequency of evasive maneuvers and per-action reaction latency within a 60-second observation window, from which a comprehensive response efficiency metric is formulated. Experimental results demonstrate that under complex road conditions characterized by multiple obstacles and grayscale transitions, the non-blocking state machine achieves a comprehensive response efficiency approximately 2.2 times that of its blocking counterpart, with average reaction time reduced by roughly 75 percent. In contrast, under simple scenarios with single or no obstacles, the performance gap between the two architectures narrows considerably. Building upon these findings, this paper further examines the applicability of blocking architectures in low-complexity tasks and high-priority atomic operations, and proposes a hybrid state machine design philosophy. The experimental methodology presented herein is hardware-agnostic and possesses strong potential for transferability to other robotic subsystems.