Adaptive weighted fuzzy ILC for personalized FES trajectory reconstruction using an anatomical prior

OBJECTIVE: The clinical translation of closed-loop functional electrical stimulation (FES) systems is hindered by three interrelated bottlenecks: the trade-off between algorithmic complexity and embedded real-time feasibility, inefficient personalization, and a lack of empirical validation of human-machine cooperative performance in stroke patients. APPROACH: We propose a physiology-informed adaptive weighted fuzzy iterative learning control (AW-FILC) algorithm, implemented on a custom STM32F407-based embedded FES platform. The core innovation is a dual-layer adaptive mechanism that incorporates subcutaneous fat thickness T_s, which is a key biophysical determinant of current attenuation measurable, as a feedforward prior. This prior is fused with real-time kinematic error feedback to dynamically modulate the learning gain and penalty factor. MAIN RESULTS: In simulations across two heterogeneous musculoskeletal models, AW-FILC achieves convergence within 5-7 iterations, with steady-state tracking accuracy of 10⁻⁵ for normalized torque and <0.13°for joint angles, outperforming PID-ILC and Type-1/Type-2 fuzzy ILC. In healthy participants (n=6), AW-FILC combined with voluntary effort achieved a trajectory tracking Pearson correlation of 0.87±0.04, substantially exceeding the pure-FES ceiling (0.80) and approaching the physiological gold standard of natural bilateral coordination (0.984±0.014). In stroke patients (n=8), AW-FILC elevated tracking correlation from 0.24 (pure open-loop FES) to 0.30±0.02. This modest numerical gain represents a qualitative shift from pathological toward normalized movement patterns. The embedded implementation achieves per-iteration execution < 2 ms on STM32F407, and the system respects rather than quantifies patient voluntary effort, aligning with activity-dependent neuroplasticity principles. SIGNIFICANCE: This work provides a computationally efficient, physiologically adaptive control core for embedded FES devices and empirically validates that AW-FILC-driven active rehabilitation bridges the gap between pathological and physiological coordination.