Gait Assist Exosuit Driven by Pneumatic Artificial Muscles: Integrating Posture Estimation and Assistance via Phase-Dependent Role Switching

Abstract

Pneumatic artificial muscles (PAMs) are widely used in soft wearable exosuits because of their high power-to-weight ratio and intrinsic compliance. For practical gait assistance, however, the device must estimate the wearer’s posture for assist timing and closed-loop control. Conventional approaches often rely on electrical sensors or vision-based systems, which can reduce wearability or require costly, environment-dependent setups. Our previous studies showed that Physical Reservoir Computing (PRC) can estimate joint motion from PAM pressure signals without on-suit sensors. However, the estimation pneumatic system has typically been implemented separately from the assistive pneumatic system, increasing hardware complexity. This paper proposes a time-shared sensing–actuation role-switching framework that integrates gait assistance and posture estimation within a single pneumatic architecture by temporally reallocating each PAM between assistive actuation and closed-system sensing according to the gait phase. In the non-assist phase, a PAM is used for estimation by observing pressure variations in a low-pressure closed configuration; in the assist phase, the antagonistic PAM is actuated to provide support. To compensate for reduced nonlinearity and memory due to simplification, we use random feature mapping (RFM) with ridge regression to estimate hip flexion–extension angle from closed-system pressure time series. We also introduce a staged control-step transition from estimation-only operation to estimation-driven assistive control, with a refractory period to suppress transient switching effects. As a proof-of-concept validation, treadmill experiments with five subjects were conducted under controlled constant-speed walking conditions. The results showed that estimation performance was sustained during assist phases, except for brief transients immediately after switching. These results support the feasibility of sensor-minimal, phase-switching PAM-based exosuits integrating posture estimation and gait assistance in a unified pneumatic system.