Personalizing impedance controllers for powered prosthetic legs is critical for accommodating individual gait biomechanics but remains challenging. Existing methods often require time-intensive human-in-the-loop exploration or constrain optimization to low-dimensional, single-joint parameter subspaces.
This paper proposes a replay-constrained simulation framework that leverages recorded human gait data to constrain the simulation, enabling high-dimensional optimization of knee-ankle prosthesis controllers without human-in-the-loop. The framework replays recorded joint kinematics and interaction forces to ensure realistic simulation conditions.
The approach allows optimization of up to 10 controller parameters across both knee and ankle joints, significantly expanding the design space compared to prior work. Simulation results demonstrate improved tracking of desired joint angles and torques compared to baseline controllers.
The framework reduces the need for extensive human subject testing, accelerating the development of personalized prosthetic controllers. It also provides a platform for exploring complex controller architectures that were previously infeasible due to optimization constraints. Source: arXiv (2607.22858).