Generating an Adaptive and Robust Walking Pattern for a Prosthetic Ankle-Foot by Utilizing a Nonlinear Autoregressive...

Area of Science:

• Rehabilitation Robotics
• Prosthetics Engineering
• Biomechanical Engineering

Background:

• Developing powered lower limb prostheses to mimic intact limb function across various speeds and terrains is a significant challenge.
• Current control strategies often employ hierarchical schemes with discrete transitions, necessitating extensive sensors, increasing computational load and cost.
• Existing methods require switching between controllers, complicating prosthetic operation.

Purpose of the Study:

• To propose a user-independent, free-mode control method for powered lower limb prostheses.
• To eliminate the need for controller switching in prosthetic systems.
• To accurately estimate foot orientation using a recurrent neural network.

Main Methods:

• A database was created using four wearable OPAL devices on seven able-bodied subjects.
• Gait data at three walking speeds were recorded for ground-level walking.
• A nonlinear autoregressive network with exogenous input (NARX) recurrent neural network (RNN) was trained to estimate sagittal plane foot orientation using shank angular velocity.

Main Results:

• The trained NARX RNN accurately estimated foot orientation across subjects and speeds on flat terrain.
• The average root-mean-square error (RMSE) was 2.1° ± 1.7°, with a minimum correlation of 86% between estimated and measured values.
• Error analysis confirmed normal distribution with a high certainty level (minimum p-value of 0.88).

Conclusions:

• The proposed user-independent method effectively estimates foot orientation for powered prostheses.
• This approach simplifies control by eliminating the need for discrete controller transitions.
• The findings suggest a promising direction for developing more intuitive and cost-effective prosthetic limbs.