Nonlinear control of a prosthetic knee with variable impedance

• Nonlinear control strategies with variable impedance enable more natural gait in prosthetic knee devices.¹
• This advances lower-limb neuroprosthetics and active prosthetic control beyond linear methods.^{1 1}

Weekly enrichment (2026-07-20)

• A 2025 Frontiers in Neurorobotics study modeled a two-degree-of-freedom prosthetic knee and introduced three robust nonlinear control laws — Integral Sliding Mode Control, Conditional Super-Twisting SMC, and Conditional Adaptive Barrier-function SMC — to handle nonlinear joint dynamics, disturbances, and modeling uncertainty.²
• In that study, controller gains were tuned with the Red Fox Optimization metaheuristic, stability was proven via Lyapunov theory, and the design was validated in MATLAB/Simulink plus hardware-in-the-loop on a Texas Instruments C2000 Delfino F28379D microcontroller; the barrier-function controller gave the best tracking accuracy and smoothest torque.²
• Continuously-varying (nonlinear) impedance control modulates joint stiffness, damping, and equilibrium angle as smooth functions of gait phase, walking speed, and incline, replacing the discrete state transitions of traditional finite-state-machine impedance controllers.^{3 4}
• Best and colleagues’ hybrid controller uses continuously-variable impedance during stance and kinematic control during swing, and in validation with two above-knee amputee participants it met or exceeded a hand-tuned finite-state-machine benchmark on most kinematic, kinetic, work, and cadence metrics.⁴
• Tuning can be made tractable by combining kinematics-informed convex optimization with PCA-based dimensionality reduction; sensitivity analysis identified stiffness as the primary driver of knee kinematics.³
• A related minimal-tuning data-driven controller let an above-knee amputee complete a sit-to-stand task about 20% faster with roughly half the asymmetry versus his passive prosthesis, and a Timed Up and Go test with only about a 10% speed decrease.⁵
• A 2026 review notes that multimodal finite-state-machine impedance controllers may require up to 140 tunable parameters and over five hours of manual tuning, motivating data-driven approaches that use real-time gait-phase estimation and model-based parameter selection.⁶
• That review also reports that fusing EMG with mechanical sensors cut steady-state locomotion-mode classification error from about 15% to about 1%, and an adaptive neural controller held a 4.03% forward-prediction error across multiple days and terrains — evidence for anticipatory, user-intent-aware prosthetic control.⁶
• Phase-variable control derived from the residual thigh angle lets a powered knee-ankle prosthesis adapt continuously across speeds and inclines using only a small parameter set, avoiding the combinatorial tuning burden of per-condition controllers.⁷

Footnotes

1. https://news.google.com/rss/articles/CBMiqgFBVV95cUxPckpBaEROSEJ5a0llRm95c25wNm41LW81MXhPZVpRcGRBUVItcGFndk1Vb0dweUNiSzRIN2FmUnJtcVllNVdFWXBEY0xURlZiMF84cFJ1UEZFMkt1WVFXUFBaR0V0Uy1sVzRkNkZFbUJpSGlVcGNWZVhtOGxJYmNRR0YyR0VxY1A0MFJSa2p3a0hOZzBOVENKNl9uSHJHSU9QT0hyY1FqenBVZnpIb0Vo?oc=5 ↩ ↩² ↩³

2. https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2025.1681298/full ↩ ↩²

3. https://pmc.ncbi.nlm.nih.gov/articles/PMC13099080/ ↩ ↩²

4. https://pmc.ncbi.nlm.nih.gov/articles/PMC10249435/ ↩ ↩²

5. https://doi.org/10.1109/iros47612.2022.9982037 ↩

6. https://link.springer.com/article/10.1186/s12938-026-01558-x ↩ ↩²

7. https://pmc.ncbi.nlm.nih.gov/articles/PMC8890507/ ↩