Design and Validation of a Lower-Limb Haptic Rehabilitation Robot

Abstract

Present robots for investigating lower-limb motor control and rehabilitation focus on gait training. An alternative approach is to focus on restoring precursor abilities such as motor adaptation and volitional movement, as is common in upper-limb robotic therapy. Here we describe NOTTABIKE, a one degree-of-freedom rehabilitation robot designed to probe and promote these underlying capabilities. A recumbent exercise cycle platform is powered with a servomotor and instrumented with angular encoders, force-torque sensing pedals, and a wireless EMG system. Virtual environments ranging from spring-mass-damper systems to novel foot-to-crank mechanical laws present variants of leg-reaching and pedaling tasks that challenge perception, cognition, motion planning, and motor control systems. This paper characterizes the dynamic performance and haptic rendering accuracy of NOTTABIKE and presents an example motor adaptation task to illustrate its use. Torque and velocity mode controllers showed near unity magnitude ratio and phase loss less than 60 degrees up to 10 Hz. Spring rendering demonstrated 1% mean error in stiffness, and damper rendering performed comparably at 2.5%. Virtual mass rendering was less accurate but successful in varying perceived mass. NOTTABIKE will be used to study lower-limb motor adaptation in intact and impaired persons and to develop rehabilitation protocols that promote volitional movement recovery.

Fig. 1.

NOTTABIKE is a one degree-of-freedom robot used to study human motor control and to deliver rehabilitation in the lower-limb. Measurements of subject endpoint kinematics and kinetics are used by a computer controller to create virtual haptic environments.

Fig. 2.

Two-stage mechanical drivetrain of NOTTABIKE. The drivetrain provides efficient power transfer between the user, who interacts through the pedals, and the industrial servomotor.

Fig. 3.

Schematic representation of the system architecture. Measurements from encoders, force sensors, and EMG sensors are read through a LabVIEW Virtual Instrument at 1000Hz and are streamed to a virtual environment loop running in (Python) Jupyter Notebook at 100Hz. The state of the robot is updated, and a command torque or velocity is calculated based on the currently selected haptic environment from the Environment Library. Experiments may be designed and executed in the Protocol Script using tools from the Trial Management Library. Outputs are then sent to the motor amplifier and visualization program to provide performance feedback to the user.

Fig. 4.

(A) Torque command accuracy to a ramp function over a 30 second trial. (B) Average torque step response from a baseline torque of 15 Nm to a target torque of 40 Nm. Rise time was determined to be 29 ms. (B inset) Pedal fixation arrangement for torque response testing. Torque was controlled by the motor’s internal circuitry only, and measured with the pedal load cell. (C) Average frequency response function to a torque chirp input baseline torque was 20Nm with 10Nm peak-to-peak magnitude. Notable features include magnitude of approximately unity and phase lag less than 29 degrees up to 10 Hz.

Fig. 5.

(A) Velocity command accuracy to a ramp function over a 30 second trial. (B) Average velocity step-response from a baseline velocity of 1.0 rad/s to a target velocity of 3.14 rad/s (30 RPM). Rise time was determined to be 36 ms. (C) Average frequency response function to a velocity chirp input. Notable features include magnitude ratio within 1dB (12%) of unity and phase lag less than 59 degrees up to 10Hz.

Fig. 6.

(A) Crank angle and torque over time during an impedance-based haptic rendering of a spring (k = 40 Nm/rad). (B) Regression between measured torque and measured crank angle for the spring of part A. (C) Crank velocity and crank torque vs. time during an impedance-based rendering of a damper (c = 15 Nm(rad/s)⁻¹) (D) Regression between measured torque and crank velocity for both impedance- and admittance-based haptic damper renderings.

Fig. 7.

Demonstration of motor adaptation to springs of different stiffnesses on NOTTABIKE in a single subject under visual feedback of performance with a clock widget (inset). 50 reaching trials were performed: 80% with the medium stiffness k = 40 Nm/rad and 10% catch trials to each of k = {10, 70} Nm/rad. Subject attempted to reach 45deg (.78rad) in the least time possible.