Human-in-the-Loop Hip Exosuit Optimization for Uphill Walking: Stride Time and Mechanical Power from Onboard Sensors

Human-in-the-loop (HITL) optimization can personalize exosuit assistance, but respirometry-based cost functions require bulky equipment, and proxy-based alternatives often require additional sensors or leave their relationship with metabolism unquantified. Our previous onboard-sensor approach optimized a single objective based on positive mechanical power (PMP). This exploratory study investigated whether stride time (ST), a gait response measured by the exosuit’s thigh-mounted inertial sensors, better represents metabolic reduction. Using grid-search data from three participants walking at 1 m/s on a 5° incline with a hip extension exosuit (six subject-day cases), we evaluated PMP, ST, and their combination by multiple linear regression, termed the multi-variable cost function (MVCF), with leave-one-subject-out cross-validation. ST predicted metabolic reduction numerically better than PMP (r = 0.63 ± 0.14 vs. 0.51 ± 0.37), and the MVCF performed best (r = 0.68 ± 0.11; lowest mean absolute error), although no difference was statistically significant. The MVCF, computed solely from onboard sensors, served as the objective of Bayesian HITL optimization in five participants. The optimized assistance significantly reduced net metabolic rate by 13.3%, 7.5%, and 5.2% versus power-off, no-suit, and biomimetic conditions, respectively. Combining device-side mechanical power with human-side gait timing may provide an onboard-sensor cost function grounded in measured metabolic data.