FSM-HSVM-Based Locomotion Mode Recognition for Exoskeleton Robot
Round 1
Reviewer 1 Report
Title: “FSM-HSVM-Based Locomotion Mode Recognition for Exoskeleton Robot”
The authors proposed a hierarchical support vector machine recognition algorithm based on finite state machine (FSM-HSVM) to accurately and stably recognize the locomotion mode recognition of exoskeleton robot. In this method, the angle information of hip joint and knee joint collected by inertial sensing units (IMUs) on the thighs and shanks of exoskeleton and the plantar pressure information collected by force sensitive resistors (FSRs) are used as input signals. This method combines the finite state machine (FSM) with the common locomotion modes to establish a mode transition framework, then the hierarchical support vector machine (HSVM) recognition model is deeply combined with the mode transition framework to recognize five typical locomotion modes and eight locomotion mode transitions in real time. The authors claim that the algorithm not only reduces the sudden change in locomotion mode recognition, but also greatly improves the recognition efficiency. In addition, some experiments were performed on six subjects separately to evaluate the recognition performance. The authors claim that the results show that the average accuracy of all motion modes is 97.106%±0.955%, and the average recognition delay rate is only 25.017% ± 6.074%. Finally, tha authors claim that this method has the advantages of small calculation amount and high recognition efficiency, and can be widely applied in the field of robotics.
General comment: This work should be reworked to enhance its quality and impact. In particular, a mix between the standard sections is present in all sections lowering the quality and the impact of the manuscript. The value of the work is not clearly explained nor discussed so interested readers can not judge about the relevance of the work. Therefore all the manuscript should be deeply reworked and improved.
Some detailed comments:
lines:”
In recent years, scholars at home and abroad have conducted in-depth researches on 40
locomotion patterns recognition and achieved a lot of achievements [5-8]. Electromyogra- 41
phy (EMG) [9,10] is one of the most important signals recognized in motion mode because 42
of its early predictive properties [11]. “
and
lines:”
Specifically, according to the Cartesian 55
product of walking speed, the intra group correlation coefficient (ICC) and Dempster 56
Shafer (D-S) data fusion theory were used for data fusion, and the hidden Markov model 57
(HMM) combined with the motion state of the previous step was used to identify the real- 58
time motion state, which the average accuracy is 95.8% “
*) These lines are not clear please expand and improve
lines:” In this paper, we propose a hierarchical support vector machine algorithm based on 76
finite state machines (FSM-HSVM). The algorithm deeply combines the finite state ma- 77
chine with common locomotion modes to establish a mode framework, then organically 78
combines this framework with the hierarchical support vector machine recognition 79
model. The data of hip angle, knee angle and plantar pressure that collected by IMUs and 80
FSRs are preprocessed as signal input for training. After that, one or all of the sub-classi- 81
fiers in HSVM model are invoked in FSM framework for real time recognition. In order to 82
verify the effectiveness and accuracy of the proposed algorithm, a lot of experiments are 83
carried out under five typical motion modes and eight locomotion mode transitions.”
*) The authors should provide a more general background about hierarchical support vector machine algorithm based on finite state machines (FSM-HSVM) for interested readers not too familiar with this topic. The authors should also present in a better way the current state the art.
Lines: “According to a large number of experimental studies [21,22], the joint angle of lower 121
limb and plantar pressure show very obvious periodic changes when people walk with ..”
*) Could the authors provide a more complete background. Only two citations seem to be too few..
lines: “Taking walking on the flat ground as an example, the toe off time is defined as the 137
starting time of a cycle. At this time, the swinging leg is at the back of the human body 138
and the hip angle of the swinging leg is near the minimum value. With the increasing of 139
swing period, the flexion angle of hip joint increases gradually. The flexion angle of hip 140
joint reached the maximum after the swing foot touched the ground into the support pe- 141
riod. With the increasing of support period, the hip flexion angle began to decrease grad- 142
ually. At the end of the support period, the hip angle reaches the minimum value. The 143
change law of knee joint is similar to that of hip joint, and the main difference is the time 144
and range of extreme value. The plantar pressure obviously shows that the heel touches 145
the bottom first and the sole touches the bottom later, and the change curve of pressure is 146
very smooth, without multiple extreme points. 147
When going up stairs and ramp, the knee angle starts to decline from the peak at the 148
end of the swing period, and rebounds for a period in the early and late support, then falls 149
back and decreases. The change range and the peak value of hip angle is larger than that 150 ,etc”
*) Within the “Materials and methods” section the authors should provide the methods used in this work not to describe the results nor comment them (all these information should be provided in “Results” and “Discussion” sections”.
Lines: “For the convenience of discussion, the following is a unified definition of the concepts 170
related to locomotion mode. The locomotion modes of human body can be divided into 171
two categories, namely stable locomotion mode and locomotion mode transition. Stable 172
locomotion mode refers to continuous motion in a single locomotion mode, and locomo- 173
tion mode transition refers to the rapid transition from the current locomotion mode to 174
another locomotion mode in the process of motion. The transition time of the locomotion 175
mode is uniformly defined as the moment when the leading foot of the previous mode 176
just leaves the ground, and the cycle from this time until the next time this foot first leaves 177
the ground in next mode is called the locomotion mode transition period. “
*) See the previous comment
2.3. Motion pattern recognition algorithm 179
2.3.1. Hierarchical support vector machine classifier (HSVM)
Figure 3. Segmentation of different locomotion modes by HSVM
*) All this paragraph (and eventually the figure) belongs to the “Introduction” section. Only the methods used in this work should be provided in detail.
2.3.2. FSM-HSVM
Figure 4. Finite state machine transition diagram of locomotion modes
*) See the previous comments
lines:” The time required for classification and recognition of HSVM and FSM-HSVM mod- 322
els is discussed in detail below. Assume that ?is the total number of identified data sam- 323
ples, ?is the proportion of flat walking data in the total number of samples (0 ≤ ?≤ 1), 324
and the number of motion mode categories is ?(?≥ 3). The recognition time of HSVM 325
model and FSM-HSVM are as follows. Etc..,
???
*) The authors should provide only what is related to “Methods” section. All the other information belongs to the “Introduction”
lines: “A total of six subjects volunteered to participate in this experiment, and the height of 372
these subjects is 1.60m~1.80m and the weight is 60kg~80kg. All participants have no phys- 373
ical diseases and are informed of all experimental procedures before the experiment. Be- 374
fore the experiment, carefully check the connection of each structural member and the 375
information acquisition system of exoskeleton to ensure that the sensor is in a standard 376
state and calibrated. When the subject wears the exoskeleton, the length of the exoskeleton 377
rod is adjusted according to the feeling of the subject. Then the subjects walked adaptively 378
for about one minute, and then kept fine-tuning to ensure that human comfort and flexi- 379
bility were in the best state. 380
Each subject wore an exoskeleton robot and carried out two kinds of experiments at 381
approximately constant speed without load. The first experiment is a stable mode exper- 382
iment under different locomotion modes. The subjects walk in five modes, including flat 383
walking (FW), up the stairs (US), down the stairs (DS), up the ramp (UR) and down the 384
ramp (DR), as shown in Figure.5. The staircase is 1.5m wide, 40cm deep, 15cm high and 385
has an inclination of about 26 °. The grade of the ramp is about 7 °, and the length is 4.5m. 386
Each exercise mode included five experiments for one subject, each lasting about thirty 387
seconds. The second experiment is the transition experiment of different locomotion 388
modes. Each experimenter walks from the flat ground to another locomotion mode, and 389
from other modes to the flat ground, with a total of eight conversion modes, i.e. FW→US, 390
US→FW, FW→DS, DS→FW, FW→UP, UP→FW, FW→DP, DP→FW. Each transition period 391
included five experiments of one subject, which lasted for thirty seconds. 392
Once enough training data sets are obtained, the FSM-HSVM model can be trained 393
offline. In order to filter the noise and interference in the data collected by the acquisition 394
system, the second-order Butterworth low-pass filter is used to filter the data at a cut-off 395
frequency of 5 Hz. Locomotion mode recognition is executed online in real time. The rec- 396
ognized pattern is recorded in the processor through text, and verified and analyzed after 397
the experiment is completed. 398
All experiments were approved by the medical and experimental animal ethics com- 399
mittee of Beijing institute of technology, and the procedures used in this study followed 400
the specified principles. “
*) This section belongs to “Methods” and not to “Results”
Figure 5. Experiments of different locomotion modes. a) Flat Walking; (b) Up the Stairs ascent; (c) 406
Down the Stairs; (d) Up the Ramp; (e) Down the Ramp.
*) See the previous comments
3.2. Experimental results and analysis
*) This is the real “Results” section. Please expand and explain in a good language.
4. Discussion
lines: “The experimental results show that the average recognition accuracy of five different 474
motion modes is 97.106% ± 0.955%. The average recognition accuracies of SVM and HSVM 475
algorithms are 90.826% ± 2.391% and 92.807% ± 1.927% respectively. According to the 476
above results, it is found that the algorithm proposed in this paper has higher recognition 477
efficiency and recognition stability, etc.”
*) Please underline the novelty of this work with respect to the current state of the art.
Author Response
Thank you very much for giving us an opportunity to revise our manuscript. We appreciate the editor and reviewers very much for their constructive comments and suggestions on our manuscript entitled “FSM-HSVM-Based Locomotion Mode Recognition for Exoskeleton Robot”. (ID: applsci-1666046).
We have studied reviewer’s comments carefully and have made revision which marked in red in the paper. We have tried our best to revise our manuscript according to the comments. Please see the attachment, and we would like to submit for your kind consideration.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors propose a hierarchical support vector machine recognition algorithm based on finite state machine to recognize locomotion mode. The overall presentation of the paper is good, and the description of the applied methods and evaluation experiments are sufficient. Some suggestions and comments are given as followings.
- ‘Long [18]’, ‘Yin [19]’, ‘Wang [20]’ … the citation style should be improved. The authors of these paper are not only one author. Similar problems should be double check throughout this manuscript.
- Although it is not the main goal of this paper, it is important that the authors describe some details on the exoskeleton functionalities, this also helps on the understanding of purpose of locomotion mode recognition.
- Line 168 Page 5, ‘Where’ should be revised as ‘where’, no space is suggested.
- Line 275 Page 8, ‘In formula (2) above,’ is suggested to revise as ‘where’. Similar problems should be revised, such as Line 345, Line 350, Line 352, etc.
- Line 329, ‘。’ should be revised.
- Line 416, ‘Table 3-5 below’ should be revised as ‘Tables 3-5 below’.
- Line 423 ‘Table.4’ should be revised as ‘Table 4’. Lines 425 and 427 have similar problems.
- [18] and [19] are the important references for this study. The major innovations should be further emphasized to improve the quality of this paper.
- There are some writing errors in the paper, which need to be corrected.
- The processing of data must be described well, but it is not clear enough in this version.
- It is mentioned that DTW algorithm is used for clustering, and the process of this algorithm should be described.
Comments for author File: Comments.pdf
Author Response
Thank you very much for giving us an opportunity to revise our manuscript. We appreciate the editor and reviewers very much for their constructive comments and suggestions on our manuscript entitled “FSM-HSVM-Based Locomotion Mode Recognition for Exoskeleton Robot”. (ID: applsci-1666046).
We have studied reviewer’s comments carefully and have made revision which marked in red in the paper. We have tried our best to revise our manuscript according to the comments. Please see the attachment, and we would like to submit for your kind consideration.
Author Response File: Author Response.pdf
Reviewer 3 Report
The article addresses an interesting topic of recognizing the gait pattern of a person to be replicated by an exoskeleton based on a combined method. It’s used a finite state machine to establish a mode transition framework and a hierarchical support vector machine to recognize five typical locomotion modes and eight locomotion mode transitions in real-time.
When collecting walking parameters are used IMU sensors on the thighs and shanks and force-sensitive resistors for plantar pressure.
It’s proposed as evaluation parameters the average accuracy of all motion, the average recognition delay rate, small calculation amount, and high recognition efficiency. Where are these parameters compared with the results of other studies?
How can the results be applied in other fields of robotics?
Please redo the section regarding the organizational structure using more details.
The lower limb exoskeleton robot is presented as a structure in another article, please insert the reference in the bibliography. If not, please include a block scheme. Figure 1 is wrong. It can be found in any Schematic diagram of the BIT exoskeleton robot. It refers only to the position and the parameters of the IMU sensors. Where is the plantar sensor? It could be useful to have only the exoskeleton without the wearer.
Figure 2 could be shown how many cycles are performed. Do these cycles have different sample numbers?
It could be interesting to compare only one cycle for each type of the different locomotion modes.
An analysis of the gait is modified by wearing the exoskeleton it could be useful.
Chapter 2.2. Analysis of motion characteristics should start with eq.1 and after that should be introduced the discussion of the results presented in figure 1
FSM should be presented in a subchapter i.e. 2.3.2. FSM. Now it’s presented in 2.3.2. FSM-HSVM on the 8th page. The text from “Finite-state…..” until “…the final state” should be moved in the FSM subchapter.
On lines 231 and 234 what is the meaning of DTW?
In eq. 2 M is not explained.
In eq 3 and 4 T(????) and ?(???−????) are not defined.
The phrase “It is certainly that FSM-HSVM has the highest recognition 339 accuracy and efficiency compared with other model recognition” it’s only declarative and not compared with any bibliographic reference.
On which bibliographic references are based recognition accuracy, confusion matrix (C which is not defined in the text before eq. 7).
The phrase “The complexity of the algorithm and the occupied hardware resources can be evaluated by the time of algorithm recognition” it’s only declarative and it’s not compared with any bibliographic reference. Where is defined the time of algorithm recognition and what is the link with the cumulative recognition time which I understand is presented in 3.2. Experimental results and analysis.
MEAN and SEM is not referenced in the text before table 3. Probably MEAN is the mean error and SEM is the standard error. The definition of the confusion matrix (in the table s) is not differently presented on page 9. The parameter cij is the mean or standard error?
Also, the tables 3,4, and 5 present the mean error or the recognition accuracy?
In line 343 how the probability is calculated.
In line 439 motion time of the algorithm probably is the running time of the algorithm
At line 450 RDR and is at line 451 DR probably are RD
The text from lines 465 to 473 should be moved to the conclusion section.
Author Response
Thank you very much for giving us an opportunity to revise our manuscript. We appreciate the editor and reviewers very much for their constructive comments and suggestions on our manuscript entitled “FSM-HSVM-Based Locomotion Mode Recognition for Exoskeleton Robot”. (ID: applsci-1666046).
We have studied reviewer’s comments carefully and have made revision which marked in red in the paper. We have tried our best to revise our manuscript according to the comments. Please see the attachment, and we would like to submit for your kind consideration.
Author Response File: 
Author Response.docx
Round 2
Reviewer 1 Report
It seems that the authors revised the work according to the comments of this reviewer.
No further comments .
