Human–robot collaboration (HRC) systems commonly rely on geometric separation, nominal timing schedules, and predefined safety thresholds to manage shared-workspace interaction. While these mechanisms are essential, they do not fully account for hesitation-related human behaviors—such as pausing, delayed clearance, corrective re-entry, or rework—that can signal continued engagement and elevate overlap risk if ignored. This paper presents an environment-stratified A/B simulation study comparing a baseline robotic handoff policy (Policy A) against a hesitation-aware policy driven by trained hesitation-state inference (Policy B). Policy A follows conventional handoff timing without explicit hesitation awareness. Policy B uses a classical hesitation model, trained in Python and executed via native MATLAB inference, to trigger conservative robot-side yielding when hesitation-related states are detected. The benchmark evaluates five validated collaborative environment classes—low-conflict open workspace, narrow assembly bench, precision insertion, inspection/rework, and shared-bin access—spanning low- and high-conflict shared-zone conditions. Results show that Policy B reduces mean overlap-risk events from 0.4660 to 0.0175 per run in high-conflict environments (approximately 96.2% reduction) and from 0.1100 to 0.0080 in low-conflict environments, with task completion time increasing by approximately 1.21 and 1.00 seconds, respectively. The safety improvement is achieved primarily through increased robot hold behavior rather than through increased human wait time. These findings suggest that human hesitation can serve as a useful coordination signal in collaborative robot control, particularly in high-conflict shared workspaces where timing uncertainty and rework are common. All results are simulation-based; physical robot validation remains a direction for future work.

Keywords

Human-robot collaboration, hesitation-aware control, robotic handoff, shared workspace and A/B simulation.