Towards natural stand-up movement support: guiding higher-dimensional muscle activation using a Lower-DOF assistive...

Sit-to-stand (STS) assistance should not only reduce effort but also preserve or shape neuromuscular activity patterns. We propose a data-driven control strategy for an assistive chair with two degrees of freedom (vertical and horizontal seat motion) to infer desired multi-muscle activation during STS. The chair is parameterized by four binary variables (fast/slow vertical and horizontal velocities, and early/late onset timing for each axis), yielding 16 control combinations. Surface EMG from eight lower-limb muscles was collected from six healthy adult males across all control combinations (10 trials per condition). We extracted hundred-dimensional EMG features by segmenting STS into four phases and computing summary statistics per muscle and phase. Four $L_1$ -regularized logistic regression classifiers were trained to infer each control variable from EMG features, enabling a classifier-based statistical mapping from target EMG features to chair control parameters. The classifiers achieved F-scores of 0.96 and 0.99 for forward and upward speed, and 0.89 and 0.82 for forward and upward timing, respectively. In an offline evaluation, the estimated control parameters inferred EMG feature patterns significantly closer to the target than non-target parameter combinations. These results suggest that low-DoF seat motion can be used to modulate higher-dimensional muscle activation patterns during STS, providing a basis for future real-time and individualized assistive control.