Adaptive Sliding Mode Path Tracking Control of Unmanned Rice Transplanter
Round 1
Reviewer 1 Report
The author developed an adaptive sliding mode controller for controlling the path tracking of unmanned rice transplanters. An RBF neural network and a PAR was applied to the sliding mode controller to approximate the disturbance, aimed to increase the stability of the steering control with uncertain disturbances in the rice field. The functionality of the algorithm was verified in simulation, and evaluated in field experiments. The theoretical analysis and the experimental results are solid. However, the paper was not well prepared, including the poor writing quality especially in the abstract and introduction sections, and the format inconsistency, which made the paper difficult to read. An additional suggestion would be to improve the field experiment design to demonstrate the advantages of the proposed algorithm.
Line 12: an arbitrary non-linear function
Line 16: checked -> evaluated
Line 18: method -> methods
Line 21: Is 10.7 deg a lateral deviation or heading deviation?
Line 29: Because of -> Due to
Line 34: “At home”?
There are too many improper words / incorrect sentences. Need to proofread before submition.
Line 166-172: The format is inconsistent.
Line 295: Need to describe the condition of the field, including the soil property, the unevenness, etc., which are relative to the experiment results.
Line 313: Is the result in Figure 11 produced by following the same path as Figure 10 did? If not, the performance differences might be caused by the soil condition difference during tracking. I recommend including no less than 3 trails for each tracking algorithm to better demonstrate the advantage of the proposed algorithm, and prevent the potential resultant performance difference caused by the inconsistent soil condition in the experiment.
Author Response
Dear reviewer #1: Thank you very much for your useful suggestions and instructive advices. We have revised our paper according to your comments:
- Q: “ The poor writing quality especially in the abstract and introduction sections, and the format inconsistency, which made the paper difficult to read. ”
A: Thank you very much for your reminding. According to your useful suggestion, we have revised the abstract and introduction sections as well as the format in manuscript and highlighted with blue color.
The detailed revision is as follows:
“an arbitrary non-linear function”in Line 12 is replaced by “any nonlinear function”ï¼›
“checked”in Line 16 is replaced by “evaluated”ï¼›
“method”in Line 18 is replaced by “methods”ï¼›
“Is 10.7 deg a lateral deviation or heading deviation ?”in Line 21
10.7 deg is heading deviation and revised in manuscript ï¼›
“Because of”in Line 29 is replaced by “Due to”ï¼›
“At home”? in Line 34:” is replaced by “domesticï¼›
The format in Line 166-172 has been revised.
- Q: “There are too many improper words / incorrect sentences. Need to proofread before submition ”
A: Thank you very much for your reminding. The improper words / incorrect sentences have been improved. The corresponding revisions in manuscript have been highlighted with blue color.
- Q: “Line 295: Need to describe the condition of the field, including the, the unevenness, etc., which are relative to the.”
A: Thank you very much for your reminding. The soil property relating to experiment results has been given in the manuscript and the corresponding revision in manuscript has been highlighted with blue color.
- Q: “Line 313: Is the result in Figure 11 produced by following the same path as Figure 10 did? If not, the performance differences might be caused by the soil condition difference during tracking. I recommend including no less than 3 trails for each tracking algorithm to better demonstrate the advantage of the proposed algorithm and prevent the potential resultant performance difference caused by the inconsistent soil condition in the experiment.”
A: Thank you very much for your reminding. The result in Figure 11 was produced by following the same path as Figure 10. During paddy field experiments, the vehicle track with depth about 25 cm was generated after vehicle travelled across the planned paths. If the vehicle runs along the same path many times, the depth of mud feet will become deeper and deeper. Due to the influence of sideslip and the restriction of the driving ability of the electric steering wheel, in the next round, it will be difficult for the vehicle to escape the track generated in the previous round of test. Hence, four groups of field tests with different panned paths were carried out to validate the effectiveness. Considering effect of the vehicle track of previous trial on tracking performance of current run, the traditional sliding control and proposed control algorithms were used alternately during each group of tests. For each group test, the vehicle followed the same planned paths. During group 1 and 3 tests, the traditional sliding mode control was applied first and afterwards the proposed control algorithms in this study was run. For the group 2 and 4, the running sequence of two control algorithm was the opposite. After four sets of tests, the lateral and heading errors were averaged and corresponding results were shown in Figure 10 and Figure 11.
Thank you again for your time and suggestions!
Author Response File: Author Response.docx
Reviewer 2 Report
The authors have reported the study on adaptive sliding mode path tracking control of unmanned rice transplanter. This is interesting. In general, the main conclusions presented in the paper are supported by the figures and supporting text. However, to meet the journal quality standards, the following comments need to be addressed.
1. Abstract: Should be improved and extended. The authors should cover the problem formulation. The novelty of the proposed model is missing. Also provided the general applicability of their model to attract general readers. Please be specific what are the main quantitative results to attract general audiences.
2. The introduction can be improved. The authors should focus on extending the novelty of the current study. Emphasize should be given in improvement of the theory/ model.
3. What is the complexity of RBF neural network architecture ? Did the authors employ any data augmentation methods before training? If so, it should be mentioned. Also, all hyperparameters (learning rate, mini-batch size, number of epochs, optimizer) and model complexity should be detailed
4. Please provide a fair weakness and limitation of the model, and how it can be improved.
5. Typographical errors: There are several minor grammatical errors and incorrect sentence structures. Please run this through a spell checker.
6. what about comparison of the result with current state-of-the art models. Did authors perform ablation study to compare with different models? Also What are the baseline models and benchmark results?
6. The authors should mention some of the state-of-the-art deep learning model that has used in agricultural applications ( see : Scientific Reports 11, 1447 (2021) https://doi.org/10.1038/s41598-021-81216-5, Comp Elect in Agri (2022), 193 106694 https://doi.org/10.1016/j.compag.2022.106694). Hence they can be discussed in the related work section.
Author Response
Dear reviewer #2:
Thank you very much for your useful suggestions and instructive advices. We have revised our paper according to your comments:
- Q: “Abstract: Should be improved and extended. The authors should cover the problem formulation. The novelty of the proposed model is missing. Also provided the general applicability of their model to attract general readers. Please be specific what are the main quantitative results to attract general audiences.”
A: Thank you very much for your reminding. According to your useful suggestion, the Abstract is improved and extended. The novelty of the proposed model is clarified, and quantitative results are presented in the manuscript. The corresponding revision is highlighted in manuscript with blue color.
- Q: “The introduction can be improved. The authors should focus on extending the novelty of the current study. Emphasize should be given in improvement of the theory/model”
A: Thank you very much for your suggestion. According to your suggestion, the introduction are revised to emphasize improvement of the theory/model. The corresponding revision is highlighted in manuscript with blue color.
- Q: “What is the complexity of RBF neural network architecture ? Did the authors employ any data augmentation methods before training? If so, it should be mentioned. Also, all hyperparameters (learning rate, mini-batch size, number of epochs, optimizer) and model complexity should be detailed?”
A: Thank you very much for your reminding. 2-5-1 RBF neural network architecture is employed. Considering the real-time performance of the algorithm, data from training set and validation set is only preprocessed by data normalization in this manuscript. Moreover, image data is not involved, data augmentation methods such as test-time augmentation, geometric transformations, meta learning and data augmentation based on deep learning are not applied. hyperparameters are detailed in section 3.2 in the manuscript and highlighted in manuscript with blue color. Mini-batch gradient descent is employed and mini-batch size is set 2048 by trial and error. The number of epochs is 10. To accelerate learning in the early stage of algorithm optimization and avoid large fluctuations in the later period, the learning rate is adjusted from 0.9 to 0.2.
- Q: “Please provide a fair weakness and limitation of the model, and how it can be improved.”
A: Thank you very much for reading our manuscript in detail. It is found that the model is relatively weak when rice transplanter is subject to sudden large slippage. The main reason is that the real-time performance of algorithm will be improved in the future research. Adaptive ability to sudden large slippage is perhaps improved by adding feed-forward velocity control unit.
- Q: “Typographical errors: There are several minor grammatical errors and incorrect sentence structures. Please run this through a spell checker.”
A: Thank you very much for your reminding. According to your suggestion, we have corrected grammatical errors by running a spell checker, the corresponding revisions in the text are highlighted in red color.
- Q: “what about comparison of the result with current state-of-the art models. Did authors perform ablation study to compare with different models? Also What are the baseline models and benchmark results?”.
A: Thank you very much for your reminding. The comparison between the proposed adaptive sliding control algorithm in this manuscript and pure tracking algorithm with adaptive forward-looking distance was carried out. During slippery paddy, the average of absolute heading deviation and lateral deviation from pure tracking algorithm are 2.1° and 7.8cm, respectively. the average of absolute heading deviation and lateral deviation from pure tracking algorithm are 1.3° and 3.2 cm, respectively. Moreover, the comparison between the proposed adaptive sliding control algorithm in this manuscript and traditional sliding control algorithm was also executed and the corresponding results were presented Figure 10 and Figure 11.
- Q: “The authors should mention some of the state-of-the-art deep learning model that has used in agricultural applications (see: Scientific Reports 11, 1447 (2021) https://doi.org/10.1038/s41598-021-81216-5, Comp Elect in Agri (2022),193 106694 https://doi.org/10.1016/j.compag.2022.106694). Hence they can be discussed in the related work section.”
A: Thank you very much for your reminding. According to your suggestion, the paper “Tomato detection based on modifed YOLOv3 framework.” have been referred and deeply enlightened. The corresponding citations in the text and highlighted in red color. Thank you again for your time and suggestions!
Author Response File: Author Response.docx
Reviewer 3 Report
This paper reports a path tracking control approach for unmanned rice transplanter. The authors need to address several issues in the paper, including the following:
1. The paper in general is very short. This can also be seen in the Introduction, which does not sufficiently discuss the background of the topic. The authors should also focus on recently published papers.
2. The authors presented the research problem and proposed a solution at the end of the Introduction. This can be further enhanced by stating the novelty/contribution of this paper when compared to previously reported papers.
3. In Section 3.1, the authors mentioned the following: “… the kinematic model can be expressed as [19-24]”. This sentence does not require citing five references.
4. The proposed control structure requires additional discussion. Moreover, the structure should be represented using standard control systems diagrams.
5. How will the proposed control approach perform when the vehicle turns using different turning strategies? Was the proposed approach tested using a straight path only? The authors may refer to the following paper to consider similar scenarios:
https://doi.org/10.3390/act11010022
6. The authors should consider using different loads and investigate the performance of the controller while doing so.
7. The novelty of the paper is limited as the topic has been previously investigated by many authors. One of the papers that discusses a similar topic is Reference 25. The authors need to carry out additional experiments and analyses.
Author Response
Dear reviewer #3:
Thank you very much for your useful suggestions and instructive advices. We have revised our paper according to your comments:
- Q: “The paper in general is very short. This can also be seen in the Introduction, which does not sufficiently discuss the background of the topic. The authors should also focus on recently published papers.”
A: Thank you very much for your reminding. According to your useful suggestion, we have extended the paper and the background of the topic is discussed in the Introduction. Moreover, the recently published papers are referred. The corresponding revision is highlighted in manuscript with blue color.
- Q: “The authors presented the research problem and proposed a solution at the end of the Introduction. This can be further enhanced by stating the novelty/contribution of this paper when compared to previously reported papers.”
A: Thank you very much for your suggestion. According to your suggestion, we have stated the novelty/contribution of this paper by comparison with previously reported papers in the Introduction and highlighted in manuscript with blue color.
- Q: “In Section 3.1, the authors mentioned the following: “… the kinematic model can be expressed as [19-24]”. This sentence does not require citing five references.”
A: Thank you very much for your reminding. According to your suggestion, we have reduced the references from five to one.
- Q: “The proposed control structure requires additional discussion. Moreover, the structure should be represented using standard control systems diagrams.”
A: Thank you very much for reading our manuscript in detail. The discussion of the proposed control structure is increased, and the structure is presented using standard control systems diagrams.
- Q: “5. How will the proposed control approach perform when the vehicle turns using different turning strategies? Was the proposed approach tested using a straight path only? The authors may refer to the following paper to consider similar scenarios:https://doi.org/10.3390/act11010022”
A: Thank you very much for your reminding. According to your suggestion, the paper “Path racking Control of an Autonomous Tractor Using Improved Stanley Controller Optimized with Multiple-Population Genetic Algorithm” have been referred and deeply enlightened. The corresponding citations in the text and highlighted in red color. The “Ω-turn” turning strategy is used in this study . The proposed control algorithm is performed during the turns of the rice transplanter. During paddy field experiments, the vehicle track with depth about 25 cm was generated after vehicle travelled across the planned paths. If the vehicle runs along the same path many times, the depth of mud feet will become deeper and deeper. Due to the influence of sideslip and the restriction of the driving ability of the electric steering wheel, in the next round, it will be difficult for the vehicle to escape the track generated in the previous round of test. During the process of rice transplanting, it is expected that turn zone of the rice transplanter is as small as possible and the transplanters do not enter unplanted areas to avoid generating vehicle track. Considering that the line spacing of seedling is 30 cm and “U-turn” is not used. Due to above restrictions and convenience for accurate tracing next line after turns , the “Ω-turn” turning strategy is selected and validated by field experiments.
- Q: “The authors should consider using different loads and investigate the performance of the controller while doing so.”.
A: Thank you very much for your reminding. The loads are mainly the seedling weight and driving resistance force from paddy soil. For above different loads, the field experiments are carried out Red guard farm of in Jiansanjiang and grain industrial park of XINGHUA City, JIANGSU Province, respectively. Experimental results show that the performance of the controller is satisfactory.
- Q: “The novelty of the paper is limited as the topic has been previously investigated by many authors. One of the papers that discusses a similar topic is Reference 25. The authors need to carry out additional experiments and analyses.”
A: Thank you very much for your reminding. Indeed, adaptive sliding mode control methods has been previously investigated by many authors and corresponding investigations are carried out for control of autonomous vehicles. To compensate for yaw rate plant variations from various implements, a kind of model reference adaptive control (MRAC) system by adapting the feed-forward yaw rate controller for unmanned tractors is presented in Reference 25. However, the aim of the proposed adaptive sliding mode variable structure control algorithm in the manuscript is to decrease the impact of uncertainty disturbance such as sideslip from field environment on the path tracking control accuracy of unmanned rice transplanter.In fact, many experimental tests are carried out, the detail steps as follows:
During paddy field experiments, the vehicle track with depth about 25 cm was generated after vehicle travelled across the planned paths. If the vehicle runs along the same path many times, the depth of mud feet will become deeper and deeper. Due to the influence of sideslip and the restriction of the driving ability of the electric steering wheel, in the next round, it will be difficult for the vehicle to escape the track generated in the previous round of test. Hence, four groups of field tests with different panned paths were carried out to validate the effectiveness. Considering effect of the vehicle track of previous trial on tracking performance of current run, the traditional sliding control and proposed control algorithms were used alternately during each group of tests. For each group test, the vehicle followed the same planned paths. During group 1 and 3 tests, the traditional sliding mode control was applied first and afterwards the proposed control algorithms in this study was applied. For the group 2 and 4, the running sequence of two control algorithm was the opposite. After four sets of tests, the lateral and heading errors were averaged and corresponding results were shown in Figure 10 and Figure 11.
Just as you mentioned, additional experiments and analyses still should be performed to validate the control performance of the control algorithm. In future research plan, the more experiments and analyses will be carried out. Thank you again for your time and suggestions
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
Although the authors address most of the reviewer's previous comments, however, points 6 and 7 are not sufficiently addressed. The authors should carefully discuss the current state-of-the-art models and suggested references, and how the current study advances the existing models.
I recommend further revision for these two points after taking into consideration the revised manuscript.
Author Response
Dear reviewer #2:
Thank you very much for your useful suggestions and instructive advices. We have revised our paper according to your comments:
1.Q: “Although the authors address most of the reviewer's previous comments, however, points 6 and 7 are not sufficiently addressed. The authors should carefully discuss the current state-of-the-art models and suggested references, and how the current study advances the existing models.I recommend further revision for these two points after taking into consideration the revised manuscript. ”
A: Thank you very much for your reminding. According to your suggestion, the paper “Tomato detection based on modifed YOLOv3 framework.” and “Real-time growth stage detection model for high degree of occultation using DenseNet-fused YOLOv4” have been referred and deeply enlightened. The corresponding citations in the text and highlighted in blue color. We have carefully discuss the current state-of-the-art models and suggested references in the manuscript. The deep learning model was not covered in this study, the current study advances the existing models was not essential.
Thank you again for your time and suggestions!
Author Response File: Author Response.docx
Reviewer 3 Report
The authors have addressed my comments.
Author Response
English changes of this manuscript is improved.