Fig 1.
The left side of the scheme describes the experimental protocol, that included: 2 exercises with point-to-point movements (Radial) and curvilinear movements (Curvilinear), carried out in 2 different modes (Free and Assisted) repeated each one 10 times (total of 40 repetitions). The right panel shows the kinematic experimental models and considered signals. The aim of the work is to investigate muscle synergies of the upper-limb in a context of human-interaction to characterize the effects of robot assistance. Tracking data was elaborated with the Vicon system and EMG with a 16 channel wireless system.
Fig 2.
The PlanArm2 prototype.
Fig 3.
A schematic representation of exercises trajectories: Radial (left) and curvilinear (right).
Fig 4.
A detailed scheme of the 3 output variables is shown.
1) percentage of assistance force; 2) Spatial error in Free and Assisted modes with respect to the Reference Trajectories; 3) EMG data analysis and synergy extraction. EMG data were first aligned, filtered and segmented into sub-tasks. Then, EMG data underwent removal of tonic components, and each channel was normalized to the maximum EMG found in all trials for that channel. Lastly, negative phasic EMG data were clipped to 0. Then, synergies were extracted with the NMF algorithm; the analysis of human-robot interaction was conducted by comparing spatial synergies of the Free and Assisted modes, for each exercise separately, with an intra-subject analysis.
Fig 5.
EMG elaboration and spatial synergies extraction.
In the left panel, the aligned and filtered EMG envelopes are reported in light gray for a representative subject, condition and task (O-NE). In light red, the ramp model for tonic activity is portrayed. The phasic EMG activity is achieved by subtracting the tonic activity from the original envelopes. The normalization and zero-clipping procedures have been applied in order to remove the negative phasic components (center panel) and apply the standard NMF method. In the center panel, we showed the extracted spatial synergies and the temporal coefficients in the same representative task. The C coefficients are here shown for all the 10 repetitions, overlapped one with the other. In the right panel, in red, the reconstructed EMG envelopes are shown for a representative subject and condition respectively. In light gray, the original EMG envelopes are reported.
Fig 6.
Mean inter-subject error and deviation in the Radial mode displayed for trajectories towards the targets and back to the center. For the Assisted mode, we also displayed the radius of the assistance tunnel in black.
Fig 7.
Mean inter-subject error and deviation in the curvilinear tasks in Free mode (upper panel) and in Assisted mode (lower panel) displayed for ¼ circle and ½ circle and for direction of execution (clockwise and anticlockwise). For the Assisted mode, the radius of the assistance tunnel is displayed in black.
Fig 8.
Summary of kinematics errors in the three modes.
left panel: Radial free and Radial assisted; middle panel: Curvilinear free ¼ circle and Curvilinear assisted ¼ circle and right panel: Curvilinear free ½ circle and Curvilinear assisted ½ circle. The mean kinematic error after repetition averaging and across directions is shown for each subject. We displayed the radius of the assistance tunnel in black.
Fig 9.
Mean inter-subject assistance and deviation in the Radial task, displayed for trajectories towards the targets and back to the center.
Fig 10.
Mean inter-subject assistance and deviation in the Curvilinear mode, displayed for ¼ circle and ½ circle and for direction of execution (clockwise and anticlockwise).
Fig 11.
Comparison of assistance level in Radial and Curvilinear movements.
The assistance level in Curvilinear ½ and Curvilinear ¼ exercises is higher than in Radial. The mean assistance level after repetition averaging and across directions is shown for each subject.
Fig 12.
A typical example of paired synergies coupled by similarity following the extraction in the Radial Free and Assisted for the subject 1.
We found that the matching score for muscle synergies (W) in the two modes is high: in our trial, assistance influenced task performance, but only a slight modification of muscle synergies was found.
Fig 13.
For a typical subject, comparison between spatial synergies, temporal coefficients, and directional tuning.
On the left, the spatial synergies of the two modes are represented overlapped (Free data have a black edge and a colored light area; Assisted data have a marked colored edge and an empty area). In the center of the Figure, the temporal coefficients of the MS are portrayed, with the same colorcode used for spatial synergies. On the right, the directional tuning of spatial synergies is portrayed.
Fig 14.
Distribution of mean synergy matching scores (Free vs Assisted).
It is portrayed in the Radial exercise (first row), Curvilinear ½ circle exercise (second row), Curvilinear ¼ circle exercise (third row). Our results show that the synergy matching is generally high. In the Radial exercise, statistical tests showed that some subjects’ synergy matching is different from others (subject 5 and subject 9 matchings are lower than subject 2’s); in the Curvilinear ¼ circle exercise, statistical tests showed that some subjects’ synergy matching is different from others (subject 2 and subject 3 matchings are lower than subject 5’s); in the Curvilinear ½ circle exercise, statistical tests showed that some subjects’ synergy matching is different from others (subject 9 and subject 10 matchings are lower than subject 8’s). The mean Synergies Matching Score for all subjects is reported for both exercises.
Fig 15.
Distribution of temporal coefficients matching scores (Free vs Assisted).
It is portrayed in the Radial exercise (first row), and Curvilinear ½ circle exercise (second row), Curvilinear ¼ circle exercise (third row). Our results show that the temporal coefficient matching is generally quite high. In the Radial exercise, in the Curvilinear ¼ circle exercise, and in the Curvilinear ½ circle exercise, no statistically significant differences between subjects were shown. The mean Synergies Matching Score for all subjects is reported for both exercises.