AJaP: An Adaptive Network for Hip Joint Angle Prediction in Assistive Walking With Continual Learning

Predicting lower limb joint angles during human walking is crucial for enhancing the control performance of assistive wearable robots. Existing studies typically use surface electromyography (sEMG) for joint angle prediction. However, this sensor is easily affected by measurement conditions (such as skin environment and assembly position) and requires complex preprocessing of the measured signals. This article proposes a method for predicting future hip joint angles using only motor encoders from exoskeleton robots. A joint angle predictor (JaP), trained on offline datasets, is introduced to capture comprehensive information through multiscale and multispan sampling (MSS-sampling) of the joint angle time series. Moreover, to overcome issues such as user variability, sensor data drift, and terrain change in real-world exoskeleton applications, an adaptive strategy based on continual learning (CL) is employed to improve the prediction accuracy of JaP, referred to as adaptive JaP (AJaP). In offline joint angle prediction for time horizons of 50, 100, and 200 ms, the proposed JaP achieved mean absolute errors (MAEs) of <inline-formula><tex-math notation="LaTeX">$\boldsymbol{[}$</tex-math></inline-formula>0.7846 <inline-formula><tex-math notation="LaTeX">$\boldsymbol{\pm}$</tex-math></inline-formula> 0.0859<inline-formula><tex-math notation="LaTeX">$^{\boldsymbol\circ}$</tex-math></inline-formula>, 1.5752 <inline-formula><tex-math notation="LaTeX">$\boldsymbol{\pm}$</tex-math></inline-formula> 0.0666<inline-formula><tex-math notation="LaTeX">$^{\boldsymbol\circ}$</tex-math></inline-formula>, and 2.5887 <inline-formula><tex-math notation="LaTeX">$\boldsymbol{\pm}$</tex-math></inline-formula> 0.0872<inline-formula><tex-math notation="LaTeX">$^{\boldsymbol\circ}$</tex-math></inline-formula><inline-formula><tex-math notation="LaTeX">$\boldsymbol{]}$</tex-math></inline-formula>, respectively. During exoskeleton-assisted walking experiments, AJaP achieved joint angle prediction MAEs of <inline-formula><tex-math notation="LaTeX">$\boldsymbol{[}$</tex-math></inline-formula>1.6132 <inline-formula><tex-math notation="LaTeX">${\boldsymbol{\pm}}$</tex-math></inline-formula> 0.2450<inline-formula><tex-math notation="LaTeX">$^{\boldsymbol\circ}$</tex-math></inline-formula>, 2.1850 <inline-formula><tex-math notation="LaTeX">$\boldsymbol{\pm}$</tex-math></inline-formula> 0.8219<inline-formula><tex-math notation="LaTeX">$^{\boldsymbol\circ}$</tex-math></inline-formula>, and 3.2091 <inline-formula><tex-math notation="LaTeX">$\boldsymbol{\pm}$</tex-math></inline-formula> 1.1393<inline-formula><tex-math notation="LaTeX">$^{\boldsymbol\circ}$</tex-math></inline-formula><inline-formula><tex-math notation="LaTeX">$\boldsymbol{]}$</tex-math></inline-formula> on level ground, ramp, and stair at 100 ms. With adaptive optimization, the prediction accuracy of AJaP improved by <inline-formula><tex-math notation="LaTeX">$\boldsymbol{[}$</tex-math></inline-formula>59.76%, 60.21%, and 64.77%<inline-formula><tex-math notation="LaTeX">$\boldsymbol{]}$</tex-math></inline-formula>, respectively, compared to directly deploying JaP.