Experimental Validation of Motor Primitive-Based Control for Leg Exoskeletons during Continuous Multi-Locomotion Tasks

Abstract

An emerging approach to design locomotion assistive devices deals with reproducing desirable biological principles of human locomotion. In this paper, we present a bio-inspired controller for locomotion assistive devices based on the concept of motor primitives. The weighted combination of artificial primitives results in a set of virtual muscle stimulations. These stimulations then activate a virtual musculoskeletal model producing reference assistive torque profiles for different locomotion tasks (i.e., walking, ascending stairs, and descending stairs). The paper reports the validation of the controller through a set of experiments conducted with healthy participants. The proposed controller was tested for the first time with a unilateral leg exoskeleton assisting hip, knee, and ankle joints by delivering a fraction of the computed reference torques. Importantly, subjects performed a track involving ground-level walking, ascending stairs, and descending stairs and several transitions between these tasks. These experiments highlighted the capability of the controller to provide relevant assistive torques and to effectively handle transitions between the tasks. Subjects displayed a natural interaction with the device. Moreover, they significantly decreased the time needed to complete the track when the assistance was provided, as compared to wearing the device with no assistance.

Keywords: bio-inspired; control; exoskeleton; orthosis; primitives.

Figures

Figure 1

General control diagram. From the detected locomotion task and kinematics, primitives are combined through the corresponding weights, generating muscle stimulations. These stimulations activate a seven muscle–tendon unit (MTU) model further generating the reference joint torques. A typical Hill-type MTU is outlined below the musculoskeletal model [adapted from Geyer and Herr (2010)]. See the text for the definition of the different acronyms.

Figure 2

The selected six primitives and their corresponding weights for the different walking cadences and for ascending/descending stairs. Primitive weights across the different walking cadences were further interpolated by second-order polynomials, displayed in black dashed lines.

Figure 3

(A) Locomotion track followed by the subjects. The starting corridor had a length of 12 m and each set of stairs was composed of 14 steps. Steps were 31 cm wide and 17 cm high. (B) Two different subjects wearing the devices: left, S1—female, 66 kg, 1.65 m; right, S2—male, 65 kg, 1.84 m. The following components are visible: active pelvis orthosis (APO), knee-ankle-foot orthosis (KAFO), and the shoe instrumented with a custom pressure-sensitive insole.

Figure 4

Averaged stimulations generated by the motor primitives during walking and ascending/descending stairs for all subjects.

Figure 5

Averaged kinematics and reference torques for all subjects during the trials with assistance (AM). For hip and knee, positive torques correspond to extension and negative ones correspond to flexion. For the ankle, positive torques capture plantar flexion and negative dorsiflexion. In the walking task, the literature reference corresponds to Winter (1991), the torque profiles being scaled by 15, 18, and 10% for the hip, knee, and ankle, respectively. In the stair ascending and descending tasks, the reference corresponds to Bradford and Winter (1988), the torque profiles being similarly scaled down.

Figure 6

Examples of phase error detection after task transitions: (A) from stair ascending to walking, (B) from walking to stair descending. The blue lines capture the actual phase, increasing from 0 to 1 (0–100%). The black lines capture the identification of the swing/stance phases (corresponding to 0/1, respectively). The alarm (red lines) captures the period where synchronization was considered to be lost. It is set to 1 when the phase error at the moment of foot strike is larger than 10% of the detected gait period. (C) Frequency in cycles/s for a representative TM trial of S3. The trial is divided in color segments representing periods of walking (blue), ascending stairs (red), and descending stairs (yellow).

Figure 7

Generated profiles in the commanded joint torques (blue) during representative task transitions for S2. (A) Stair ascending to walking. (B) Walking to stair descending. Black lines are constructed from reference profiles found in the literature [Winter (1991) for walking and Bradford and Winter (1988) for stair maneuvers]. These torques were scaled according to S2’s weight and walking cadence, and to the amount of delivered assistance (see Section “Participants” and “Experimental Protocol”). The red solid vertical lines capture the moment where the locomotion task change was detected. The red dotted lines capture the moment of right toe-off just before this task change was detected.

Figure 8

(A) Average and standard deviations of joint trajectories across the seven participants for hip, knee, and ankle joints for the three different tasks: walking (Winter, 1991) and stair ascending and descending (Bradford and Winter, ; Riener et al., 2002). (B) Median, 25th, and 75th percentiles of range of motion (ROM) for walking and ascending/descending (SA and SD) stairs across the seven participants. The different panels report the hip, knee and ankle ROM. The figure reports the patterns in the unassisted (TM, blue) and assisted (AM, red) conditions. For the hip and knee, values above 180° capture extension and below 180° flexion. For the ankle joint, values above 90° represent plantar flexion and below 90° dorsiflexion.