Muscles Based Controller Applied to an Active Ankle Prosthesis

 Introduction

Transtibial amputation can alter considerably the quality of life. Passive prosthesis (without energy supply) can restore basic locomotion but lacks the ability to produce healthy-like gait support. Indeed, the ankle joint needs high power peaks to propel the whole body forward during ankle pushoff. Without this power, an increase in metabolic energy consumption might appears due to muscles and joints fatigue. A motorised  ankle prosthesis is therefore necessary to restore natural gait. This project aims at designing an active artificial ankle which mimick biological functions. Its mechanical conception targets a minimization of electrical energy consumption with the use of a serie elastic actuator.  Both simulations and real tests are under development to setup bio-inspired reflexes able to adapt changing walking speed.

Mechanical design

Series elastic actuators are very popular in rehabilitation robotics. Among other advantages, elastic elements between the actuator and the load permit to store and release energy during the task completion, such that the energy balance is improved and the motor power peak is decreased. In rhythmic tasks like walking, this reduces to design the spring stiffness such that it works at resonance.

For this aim, we propose a geometric model of a conceptual prosthesis in the sagittal plane. This model was directly inspired by the series of the SPARKy devices developed by Sugar et al. [1] The active movement is transmitted to the ankle via a ball screw through one spring. The constitutive geometric parameters (i.e. R1, R2, etc...), such as the springs stiffness were optimized to reduce the motor power peak with an objective function which is the maximum power peak supply by the motor.

During the early stance phase, the motor loads the spring, which is further loaded during the load acceptance phase. During the late stance phase, the spring releases its energy (giving rise to a power peak close to the one of the healthy ankle), while power provided by the motor is very small. Without the spring, the motor would have to reproduce the power profile of a healthy ankle (peak of 170W). With the spring, the peak power is only of about 60W. In sum, the maximum power peak in the motor is decreased by a factor 3 [2].

     

Controller

Neuromuscular models of the human lower limb closely reproduce biological features of walking gait, like joint kinetic and kinematic. Using these models in prosthesis to restore human-like ankle behavior has recently shown promising results [3]. Simulating the behavior of muscles and tendons as well as spinal reflexes to mimick ankle extensor muscle groups, this model is able to adapt to changing slopes.

Starting from these previous works, we want to design an ankle controller able to adapt the prosthesis behavior to changing walking speed. Adaptive Oscillators (AOs) [4] are a set of differential equations able to learn any cyclic signal and to extract its caracteristic phase, frequency, amplitude and offset. Using this mathematical tool to extract the walking pace allows to fine tune sensory-motor reflexes of the neuromuscular model. Furthermore, AOs can be used as a feedforward motor command to muscle stimulation, or to anticipate swing/stance transition.

Simulations of the human-prosthetic interaction in Robotran are used to test such models and to find good sets of parameters. After validation in simulation, tests on the real prosthesis are realised. Both simulations and tests are ongoing.

 

 

References

[1] Bellman et al., Sparky3: design of an active robotic ankle prosthesis with two actuated degrees of freedom using regenerative kinetics, Biomedical Robotics and biomechatronics, pp.511-516, 2008

[2] Everarts C. et al., Variable Stiffness Actuator Applied to an Active Ankle Prosthesis: Principle, Energy-Efficiency, and Control, International Conference on Intelligent Robots and Systems, 2012

[3] Eilenberg M.F. et al., A neuromuscular-model based control strategy for powered ankle-foot prostheses, IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 18, NO. 2, 2010

[4] Ronsse R. et al., Real-time estimate of velocity and acceleration of quasi-periodic signals using adaptive oscillators, IEEE transactions on robotics, 2013

 Contact persons

Christophe Everarts, Steve Berger, Renaud Ronsse

©2017 Robotran