Speed-Dependent Variation of Optimal Prediction Horizon in a Bio-Inspired Model Predictive Control Framework for Slow Bipedal Walking

Abstract

Slow bipedal walking is commonly observed in older adults and patients with mobility impairments, who are among the primary users of lower-limb exoskeletons and related assistive systems. Model predictive control (MPC) is attractive for such applications because it optimizes control over a finite horizon while handling multivariable dynamics and constraints. However, most MPC studies on bipedal walking have focused on normal gait and rarely examined how the prediction horizon affects performance. This paper investigates the speed-dependent variation of the optimal prediction horizon for slow bipedal walking within a bio-inspired control framework. In the framework adopted here, the human body is represented by a compass model and neural control is represented by a linear time-varying MPC (LTV-MPC). Numerical simulations first compare the predictive controller with a proportional-derivative (PD) controller under a slow-walking condition, showing substantially smaller tracking errors and more periodic torque profiles within the prescribed constraints. Additional simulations for slow walking speeds from 0.1 to 0.6 m/s show a characteristic U-shaped relationship between tracking error and prediction horizon, together with a systematic increase in the optimal horizon with speed. These results suggest that speed-dependent selection of the prediction horizon is an important design consideration for MPC-based assistance in slow bipedal walking.