A Replay-Constrained Simulation Framework for Personalization of Powered Knee--Ankle Prosthesis Controllers

2026-07-24

AI Summary

Existing personalized impedance control methods for powered knee-ankle prostheses struggle to accommodate individual gait biomechanics and either rely on time-consuming human-in-the-loop tuning or are confined to low-dimensional parameter spaces. This work proposes a replay-constrained simulation framework that replicates prosthesis dynamics in MuJoCo and bypasses complex neuromuscular modeling by replaying individual hip kinematics and ground reaction forces. Within this framework, deep reinforcement learning simultaneously optimizes phase-dependent stiffness, damping, and equilibrium angles for both joints. To our knowledge, this is the first approach enabling high-dimensional, model-free personalization of prosthesis controllers, supporting expressive parameterizations such as neural networks. In experiments with three transfemoral amputees, simulated and hardware performance showed strong agreement (Pearson’s r = 0.96–0.997), with optimal policies consistently ranked among the top five predicted by simulation, yielding 42%–59% improvements in biomimetic reward over baseline controllers.

Abstract

Personalization of impedance controllers for powered prosthetic legs is critical to accommodating individual gait biomechanics but remains challenging. Existing methods rely on time-intensive human-in-the-loop exploration and/or constrain optimization to low-dimensional, single-joint parameter subspaces. Sim-to-real transfer has enabled high-dimensional locomotion control for legged robots, but in assistive device control the human partner remains un-modelable. We present a replay-constrained simulation framework: a MuJoCo-based simulator reproduces prosthetic knee-ankle dynamics while replaying recorded hip kinematics and feedback-based ground reaction forces from individual walking data, bypassing the need to model complex human neuromuscular control mechanisms. We demonstrate the framework with a deep reinforcement learning policy that personalizes phase-dependent stiffness, damping, and equilibrium angle at both joints simultaneously, maximizing a biomimicry-based reward computed solely from onboard prosthesis measurements. Experiments with three participants with transfemoral amputation during level-ground walking at 0.8~m/s demonstrate strong simulation-to-hardware predictive validity (Pearson $r=0.96$--$0.997$). The best-performing policy on hardware was consistently predicted within the top five simulation policies for all participants. The learned controllers improved overall biomimicry rewards by 42--59\% relative to the unpersonalized baseline. The framework supports scalable high-dimensional personalization of powered prosthetic legs and is amenable to extension to higher-dimensional controller parameterizations such as neural-network controllers.

Problem

• personalization
• powered prosthesis
• impedance control
• gait biomechanics
• sim-to-real transfer

Innovation

• replay-constrained simulation
• personalized prosthesis control
• deep reinforcement learning
• sim-to-real transfer
• biomimicry-based reward