Fig 1.
HapticMaster robot was configured to adapt its environment based on the detected muscle fatigue. EMG measured from upper limb muscles was used to detect the fatigue. The adaptive strength training algorithm forms a closed loop, as shown. Kinematic measurements, detected fatigue, reported fatigue, and demographic measurements were also analysed off-line.
Fig 2.
The experimental setup included HapticMaster robot, visual guidance and animated background. The front-end of the rowing boat was shown on the LCD display in front of the participant. The rowing environment was embedded with audio cues and haptic sensation of underwater viscosity.
Fig 3.
The EMG acquisition device with g.USBamp amplifier, three bipolar electrodes and a ground electrode.
Fig 4.
EMG electrodes are connected to three upper limb muscle locations (Biceps Brachii (BB), Anterior Deltoid (DLTF), and Middle Deltoid (DLTM)). The ground electrode was connected to a bony area near the elbow.
Fig 5.
Description of the different stages of the experiment protocol.
Table 1.
Participant table with gender, age and experiment groups.
Each participant was assigned to different groups (Control 1, Control 2, and Intervention) randomly. There were 17 male and 13 female participants of at least 18 years old.
Fig 6.
Simulink model for EMG data acquisition.
The model has an on-line signal processing algorithm that performs fatigue detection for each muscle simultaneously. The detected fatigue state was communicated to the HapticMaster control algorithm.
Fig 7.
Algorithm for robotic adaptation.
The flow chart of the adaptation algorithm for Intervention group participants.
Fig 8.
Progress of EMG median frequency.
Median frequency in a typical Intervention group participant (Subject 20), who received adaptive robotic assistance based on the detected muscle fatigue using EMG features. The doted regions represent a significant decrease in median frequency, which resulted in the detection of fatigue.
Fig 9.
The fatigue flags in a typical Intervention group participant (Subject 20), who received adaptive robotic assistance based on the detected muscle fatigue. The doted regions represent the detection of fatigue in the DLTM muscle, which decided the final state of fatigue.
Fig 10.
Task difficulty in Control-1 group.
The progress of task difficulty in Control-1 participants, who received 30 seconds break period after each trial of 1-minute duration before the MVC+ increment. This group did not receive any robotic adaptation based on muscle fatigue.
Fig 11.
Task difficulty in Control-2 group.
The progress of task difficulty in Control-2 group participants, who received a manual robotic adaptation based on the subjective fatigue reported.
Fig 12.
Task difficulty in Intervention group.
The progress of task difficulty in Intervention group participants who received an automatic robotic adaptation based on the detected fatigue using EMG features.
Table 2.
Comparison of medians of damping coefficient across the three subject groups.
Intervention group had the least value compared to Control 1 and Control 2 groups.
Table 3.
Analysis of variance (ANOVA)for the task performance measures.
Fig 13.
Box plots showing the duration of the experiment in the three groups of participants.
Fig 14.
Box plots for number of repetitions.
Box plots showing the number of repetitions of the rowing task in the three groups of participants. The Intervention group could do more task repetitions due to the auto-adaptation of the task difficulty during the progressively challenging exercise.
Fig 15.
Box plots for speed of repetitions.
Box plots showing the rate of task repetitions (repetitions/minute) of the rowing task in the three groups of participants.
Fig 16.
Box plots for time-to-fatigue.
Box plots showing the time taken to reach the first reported state of fatigue in the three groups of participants.
Table 4.
Difference between the the “time to reported-fatigue” and “time to detected-fatigue” in seconds using different thresholds for the fatigue detection algorithm.
Time-gap was calculated only for those subjects who had instances of both reported and detected fatigue. For the other participants, the time-gap could not be calculated due to the absence of a reported-fatigue because of the reasons mentioned above. Also, a few other subjects had a negative time-gap due to reporting fatigue prior to automatically detecting fatigue.
Table 5.
Correlation statistics for Intervention group.
Spearman’s correlation between participant demographic data and the experiment output measurements.
Table 6.
Correlation statistics for Control 1 group.
Spearman’s correlation between participant demographic data and the experiment output measurements.
Table 7.
Correlation statistics for Control 2 group.
Spearman’s correlation between participant demographic data and the experiment output measurements.
Table 8.
Comparing the medians across the three participant groups for the work done against resistive forces using non-parametric (Kruskal-Wallis) test.
Table 9.
Medians of work done across the three groups of participants.
Work is presented here in the units of N.cm.
Table 10.
Effect size estimate calculated using Eq 5.