Agency Calibration in Adjustable Autonomy Systems for Human-Robot Collaborative Manufacturing

Abstract

This study investigates agency calibration as a control principle for adjustable autonomy systems in human-robot collaborative manufacturing. The proposed framework integrates perceived agency, assigned autonomy, task demand, operator workload, robot confidence, error risk, decision transparency, and override quality into a measurable calibration model. Four autonomy conditions were evaluated: fixed autonomy, randomly adjusted autonomy, shared-control baseline, and agency-calibrated autonomy. The results show that agency-calibrated autonomy achieved the strongest task accuracy at 91.8%, outperforming shared control at 88.8%, fixed autonomy at 86.3%, and random adjustment at 85.7%. It also produced the highest defect avoidance rate at 90.6% and the shortest completion time at 128.4 seconds. Human-factor results showed lower workload under calibrated autonomy, with a workload score of 3.7 compared with 5.6 under fixed autonomy and 5.9 under random adjustment. Perceived control reached 5.9, decision confidence reached 6.1, and response latency decreased to 1.96 seconds. Transition analysis showed that calibrated autonomy reduced failed transitions to 0.38 and improved override appropriateness to 0.86. Explanation analysis further indicated that integrated explanations produced the highest decision acceptance at 84.7% and reduced false overrides by 34.8%. The integrated collaboration effectiveness score reached 0.88, exceeding shared control at 0.77, fixed autonomy at 0.69, and random adjustment at 0.61. These findings demonstrate that effective autonomy in collaborative manufacturing depends not on maximizing robotic independence, but on aligning robotic authority with human agency, task risk, and contextual production demand.