Lightweight Unilateral Knee Exoskeleton with IMU-Based Activity-Adaptive Torque Estimation

Abstract

This paper presents a lightweight unilateral knee exoskeleton integrated with a dual-inertial-measurement-unit (IMU)-based activity-adaptive knee torque estimation method for daily locomotion assistance. The proposed exoskeleton adopts a compact unilateral thigh-shank structure, a knee-side actuation module, modular human-machine fixation components, and integrated IMU mounting positions. The prototype has a total mass of approximately 1 kg and is designed as a portable platform for future closed-loop proportional assistance. To reduce sensor complexity and improve wearability, the estimation framework uses only triaxial acceleration and triaxial angular velocity signals from unilateral thigh and shank IMUs. A hierarchical activity classification network first recognizes static standing, level-ground walking, ramp ascent, ramp descent, stair ascent, and stair descent. Then, an activity-adaptive torque estimation network predicts the biological knee torque by incorporating activity-dependent feature weighting. Leave-one-subject-out validation was conducted on a public multi-activity locomotion dataset containing nine subjects. The proposed method achieved an average classification error of ${1. 3 5 \%} \pm {1. 4 9 \%}$ over six activities. For knee torque estimation over five dynamic activities, the method achieved an average normalized root mean square error of ${1. 6 3 \%} \pm {0. 2 3 \%}$ and an average coefficient of determination $\mathrm{R}^{{2}}$ of ${0. 7 2} \pm {0. 0 7}$. Continuous multi-terrain evaluation further indicates that the estimated torque can follow the main trend of the biological knee torque under different locomotion modes. These results suggest that the combination of a lightweight unilateral mechanical platform and a dual-IMU-based biological torque estimator can provide a feasible high-level torque reference for future portable knee exoskeleton assistance.