Review Reports

Round 1

Reviewer 1 Report

The problem investigated is important and highly relevant. First, this paper introduces the design of the lower limb exoskeleton and the human motion-sensing measurement device. The sensing measurement device can record the movement trajectory of the wearer's lower limbs, and its role is to prepare for the control of the powered exoskeleton. Secondly, a neural network and Long-Short Term Memory (LSTM) machine learning model method is proposed to predict the actual movement trajectory of the human body's lower limbs. The authors concluded that the experimental results show that the LSTM model can accurately predict the gait trajectory with an average mean square error of 0.5096. This method has practical significance for the control of the trajectory of the dynamic exoskeleton. I believe that the topic taken by the authors is interesting and current. In despite of that, the paper cannot be published unless a minor revision is done, addressing the following points:

• It's necessary to improve the introduction. You should explain better what you mean by "There are many gait detection methods, such as gait detection based on plantar pressure, gait detection based on IMU signal, and gait detection based on EMG signal [9-12]. Data processing methods include Threshold Methods, Machine Learning Methods, etc. [13]”. It's necessary to describe the context more broadly.
• It would be interesting to see the impact of different factors, on the methodology. Such analysis would further help to assess the effectiveness of the method (to consider including in future studies). I suggest that some information about the limitations and the strengths of this paper should be added. I would recommend a more sentences regarding directions for future research. I feel a sentence to engage the readers for what research should be done in the future, would enhance the closing remarks of your paper. Overall the paper is good and has potential to contribute to the field of knowledge.

 

 

Author Response

Thank you for your comments on our paper. According to your suggestions, We have rectified some related context in our previous draft. Based on your comments, the detailed point-by-point responses are listed below.

 

To reviewer #1

(1)It's necessary to improve the introduction. You should explain better what you mean by "There are many gait detection methods, such as gait detection based on plantar pressure, gait detection based on IMU signal, and gait detection based on EMG signal [9-12]. Data processing methods include Threshold Methods, Machine Learning Methods, etc. [13]”. It's necessary to describe the context more broadly.

As the Reviewer suggested, we have rectified the related content. The relevant content is as follows：

There are many gait detection methods. Such as in [9], the plantar force can be predicted using only data measured by inertial measurement unit sensors. In [10], Surface Electromyography (sEMG) is used to reflect the motion intention of the human body ahead of muscle action. In [11], the foot trajectory estimated by an IMU was used to derive the inclination grade of the terrain that they traverse to identify the locomotion mode. In [12], human motion is analyzed, and the accuracy of predicting lower extremity joint angular position based on cane motion is investigated. And so on. Data processing methods include Threshold Methods, Machine Learning Methods, etc. The simplest calculation method in gait detection is the threshold value. There are different thresholding algorithms as a set of value rules to determine specific characteristics of gait phases or events. ML methods are used to classify gait phase offline data and real-time data. Different machine learning methods, such as HMMs, NN models, DLNNs, and CNNs, have been used for gait stage recognition. [13].

(2)It would be interesting to see the impact of different factors, on the methodology. Such analysis would further help to assess the effectiveness of the method (to consider including in future studies). I suggest that some information about the limitations and the strengths of this paper should be added. I would recommend a more sentences regarding directions for future research. I feel a sentence to engage the readers for what research should be done in the future, would enhance the closing remarks of your paper. Overall the paper is good and has potential to contribute to the field of knowledge.

As the Reviewer suggested, I have added the information about the limitations and strengths in this paper. The final additions in the text are as follows：

Despite the many advantages of experimentation, there are still some disadvantages. For example, the experimental environment is single. Gait data collection and prediction can be performed later through different environments, such as stairs, slopes, and ruins. In the experiment, only the change of the joint angle is collected, and in the future, a variety of sensor data can be comprehensively considered for prediction. Finally, I hope that interested readers will join me in the study of using physical information of the human body to predict body posture.

Reviewer 2 Report

This paper studies the prediction of motion trajectories of lower limb exoskeletons.

The convolutional neural network is usually used for target recognition (classification), and in this paper it is used for gait prediction, which is innovative.

(1) The application scenarios of the proposed method can be clearly stated. For exoskeletons for rehabilitation training, the gait has good repeatability and is easy to predict. If the exoskeleton is used for daily walking, including upstairs, downhill, etc., when the gait is not highly repeatable, what is the prediction effect?

(2) When training a convolutional neural network, it is usually necessary to mark with rectangular frames. How is the training set in this paper created?

Author Response

Thank you for your comments on our paper. According to your suggestions, We have rectified some related context in our previous draft. Based on your comments, the detailed point-by-point responses are listed below.

 

To reviewer #2

(1) The application scenarios of the proposed method can be clearly stated. For exoskeletons for rehabilitation training, the gait has good repeatability and is easy to predict. If the exoskeleton is used for daily walking, including upstairs, downhill, etc. when the gait is not highly repeatable, what is the prediction effect?

The gait prediction scenario in this paper is that a healthy human walks normally on flat ground. In the follow-up, gait predictions for scenarios such as upstairs and slopes will be carried out successively, and their prediction effects will be evaluated.

(2) When training a convolutional neural network, it is usually necessary to mark with rectangular frames. How is the training set in this paper created?

The training set creation rules in this article are as follows:

A set of standard human walking data was obtained from the CMU motion database, and the hip, knee and ankle joint data of the left and right sides of the human body were extracted, respectively. Then perform data processing, taking the first 60% of the dataset as the training set and the last 40% as the test set. The first data to the penultimate data are used as predictors in the training set data, and the second data to the last data is used as the corresponding data for training. Then we can predict the last 40% of the data through the trained function.