An Anthropometry-Based Personalization of Powered Knee Prosthesis for Metabolic Efficiency

Traditional tuning methods of assistance parameters rely on the experience of human experts but often fail to achieve optimal performance. Human-in-the-loop optimization improves parameter selection but requires extensive in-lab testing. In this study, we rigorously tested two control parameters, early stance knee flexion angle (5° to 12°) and swing initiation timing (55% to 65% of the gait cycle), with ten individuals with transfemoral amputation using a commercially available robotic prosthetic knee, Össur Power Knee, and a passive foot, Pro-Flex LP. We measured energy expenditure, joint work, and user preferences during treadmill walking. Results showed a 15.6% reduction in metabolic rate with stance flexion decreasing from 12° to 5° (p<0.05). User preferences favored lower stance flexion and personalized swing initiation. Personalized-best settings reduced the metabolic rate by 4.1% (stance flexion) and 9.8% (swing initiation) compared to the best-on-average settings (p<0.05). These reductions were also significant when compared to the device default and clinically tuned settings (p<0.05). We proposed an offline learning approach using anthropometric, gait, and prosthesis-related data to estimate optimal settings, yielding a 7.1% reduction in metabolic rate (p<0.05). Our results suggest that this approach achieves comparable energy efficiency without lengthy experiments, enabling automatic parameter tuning with initial measurements.