Adaptive Ultrasound-Based Tractor Localization for Semi-Autonomous Vineyard Operations
Round 1
Reviewer 1 Report
The paper proposes a very interesting work about an adaptive localization system for semi-autonomous navigation of agricultural tractors in vineyards, using ultrasonic sensors and an algorithm based on the EKF to estimate the distance between the left and right vine rows.
The proposed methodology results very interesting. However, some revisions could further improve the work.
The Introduction section is filled with relevant references, yet I think it would be interesting to add a paragraph about the smart agriculture phenomenon.
To this purpose, some references can be added to the section, i.e.: Prasad, D.; Singla, K.; Baggan, V. System Model for Smart Precision Farming for High Crop Yielding. J. Comput. Theor. Nanosci. 2019, 16, 4406–4411. Romeo, L.; Petitti, A.; Marani, R.; Milella, A. Internet of Robotic Things in Smart Domains: Applications and Challenges. Sensors 2020, 20, 3355.
The graphs in the paper are very comprehensive, but I think they would be more clear using different colors.
For instance, in Figure 16 the shown distances are very confusing, and they would be certainly more clear if they were represented with four different colors. The same consideration can be done for the other figures in the paper.
In lines 154-155 there is a typo. The sentence here is: "Once the two rows are clustered. The RANSAC algorithm [17] fits a straight line describing the two rows.", but I think it should be: "Once the two rows are clustered, the RANSAC algorithm [17] fits a straight line describing the two rows."
Author Response
Dear Reviewer,
We wish to thank you and the reviewers for the constructive comments, which helped improve the paper quality. We are glad to submit a revised version of the manuscript and a detailed response to the review.
We did our best to address all the comments.
The numbering of equations, references and sections in this document refer to the revised version of the paper. To simplify the reviewers' effort, new and modified sections are highlighted in blue in the revised document.
Best regards,
The Authors
Point 1: The paper proposes a very interesting work about an adaptive localization system for semi-autonomous navigation of agricultural tractors in vineyards, using ultrasonic sensors and an algorithm based on the EKF to estimate the distance between the left and right vine rows.
The proposed methodology results very interesting.
Response 1: We thank the reviewer for the time they devoted to reading our paper and their positive opinion
Point 2: The Introduction section is filled with relevant references, yet I think it would be interesting to add a paragraph about the smart agriculture phenomenon.
To this purpose, some references can be added to the section, i.e.: Prasad, D.; Singla, K.; Baggan, V. System Model for Smart Precision Farming for High Crop Yielding. J. Comput. Theor. Nanosci. 2019, 16, 4406?4411. Romeo, L.; Petitti, A.; Marani, R.; Milella, A. Internet of Robotic Things in Smart Domains: Applications and Challenges. Sensors 2020, 20, 3355.
Response 2: We thank the reviewer for the suggestion. We have added the relevant literature and comments on precision farming.
Point 3: The graphs in the paper are very comprehensive, but I think they would be more clear using different colors For instance, in Figure 16 the shown distances are very confusing, and they would be certainly more clear if they were represented with four different colors. The same consideration can be done for the other figures in the paper
Response 3: We have regenerated all graphs in color to improve readability.
Point 4: In lines 154-155 there is a typo. The sentence here is: "Once the two rows are clustered. The RANSAC algorithm [17] fits a straight line describing the two rows.", but I think it should be: "Once the two rows are clustered, the RANSAC algorithm [17] fits a straight line describing the two rows.
Response 4: This typo, and several others, have been corrected. Thank for the careful review.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Authors in their paper present a cost-effective localization system for (semi-) autonomous navigation of agricultural tractors in vineyards. The algorithm estimates the distance from the left vine row and the incidence angle by means of ultrasonic sensors. The estimation algorithm is based on an Extended Kalman Filter and a data-selection step. The EKF is based on a model that also considers, as a state, the distance between the left and right rows. This makes the estimation more robust to protruding branches and holes in the rows. The adaptive data-selection step further robustifies the estimation by discarding wrong measurements. This step discards measurements that pass through the rows or hit branches far from the main row. According to their experimental results, authors find a distance RMSE of 16 cm and an angular RMSE of 2.6 degrees. They present an interesting approach for localization, taking into consideration that low cost but also low quality and accuracy ultrasonic sensors are used in measurements. Some comments and remarks are the following:
__ Grammatical and syntactic errors exist in the text. Some phrases are not clearly written. Please correct them all.
__When someone presents a system like this, the accuracy and the precision, as well as different features of the system depends on the sensors mounted on. Please add in the paper more details concerning the sensors and if it is possible the sensor models for: a) The “Wheel velocity V sensor” (Is it an encoder or something else?) b) The “front steering wheel sensor” c) The “set of 12 ultrasonic sensors”. d) The “3D LIDAR sensor”.
__ “production grade sensor”. Do you mean commercial-grade sensors or something else?
__During sampling by means of ultrasonic sensors (lines 171-173) the tractor is moving or not? Because in the first case (as the tractor is moving) measurements of different sensors depend on this movement and /or the movement velocity of the tractor. In the second case, the tractor does not smoothly navigate. Moreover, ultrasonic sensors have problems with smooth surfaces, reflections and so on. How authors handle these problem. Is sufficient the improvement that the EKF algorithm offers combined with the adaptive measurements? Is their proposed approach, finally, efficient to cope with all these problems? Please explain.
__In semi-autonomous or autonomous systems, to achieve reliable navigation, vehicle reliable localization is desirable in any case and it cannot be connected with the season or the period of production cycle. However, authors present different distance and angle RMSE depending on “the window” and the seasonal condition. How can authors improve this behavior? Please explain.
Author Response
Dear Reviewer,
We wish to thank you and the reviewers for the constructive comments, which helped improve the paper quality. We are glad to submit a revised version of the manuscript and a detailed response to the review.
We did our best to address all the comments.
The numbering of equations, references and sections in this document refer to the revised version of the paper. To simplify the reviewers' effort, new and modified sections are highlighted in blue in the revised document.
Best regards,
The Authors
Point 1: Authors in their paper present a cost-effective localization system for (semi-) autonomous navigation of agricultural tractors in vineyards. The algorithm estimates the distance from the left vine row and the incidence angle by means of ultrasonic sensors. The estimation algorithm is based on an Extended Kalman Filter and a data-selection step. The EKF is based on a model that also considers, as a state, the distance between the left and right rows. This makes the estimation more robust to protruding branches and holes in the rows. The adaptive data-selection step further robustifies the estimation by discarding wrong measurements. This step discards measurements that pass through the rows or hit branches far from the main row. According to their experimental results, authors find a distance RMSE of 16 cm and an angular RMSE of 2.6 degrees. They present an interesting approach for localization, taking into consideration that low cost but also low quality and accuracy ultrasonic sensors are used in measurements.
Response 1: We thank the reviewer for the time they devoted to reading our paper and their positive opinion
Point 2: Grammatical and syntactic errors exist in the text. Some phrases are not clearly written. Please correct them all.
Response 2: We have proofread the entire paper and corrected all typos and syntactic errors.
Point 3: When someone presents a system like this, the accuracy and the precision, as well as different features of the system depends on the sensors mounted on. Please add in the paper more details concerning the sensors and if it is possible the sensor models for: a) The ”Wheel velocity V sensor” (Is it an encoder or something else?) b) The ”front steering wheel sensor” c) The ”set of 12 ultrasonic sensors”. d) The ”3D LIDAR sensor”
Response 3: We thank the reviewer for the comment. We have detailed the specs of all the referred sensors. Please note that for the vehicle velocity and steering angle sensor, being standard sensors, we do not have the detailed specifications..
Point 4: ”production grade sensor”. Do you mean commercial-grade sensors or something else?
Response 4: We apologize for not being clear. We meant that those sensors are installed in the production vehicle. We have rephrased the statement for clarity
Point 5: During sampling by means of ultrasonic sensors (lines 171-173) the tractor is moving or not? Because in the first case (as the tractor is moving) measurements of different sensors depend on this movement and /or the movement velocity of the tractor. In the second case, the tractor does not smoothly navigate. Moreover, ultrasonic sensors have problems with smooth surfaces, reflections and so on. How authors handle these problem. Is sufficient the improvement that the EKF algorithm offers combined with the adaptive measurements? Is their proposed approach, finally, efficient to cope with all these problems? Please explain.
Response 5: The reviewer raises an interesting issue. The algorithm can estimate the distance and the incidence angle while the tractor is moving. It does not need to stop to finish the entire scanning of the sensor. One has two options two deal with the fact that the onboard electronics need 0.6 second to finish the entire cycle of all 12 sensors;
Option 1. Have the kalman filter work with a sampling time of 0.6 seconds and a time invariant model with 12 output equations. This solution has two issues. The first is the one put forward by the reviewer: we would loose measurement synchronicity. The second has to do with the fact that a sampling time of 0.6 second could be too low for control purposes. Option 2. Use each measurement as soon as it is available and have the Kalman filter run at a 50 ms sampling time. In this way, the Kalman filter is implemented on a model with a single measurement equation that is time-varying. The equation that describe the sensor measurement is determined by (4) based on the active sensor. The advantage of this approach is that the issue with the synchronicity does not arise as each measurement is employed at the very time it is measured and the fact that the vehicle moves between measurements is correctly accounted by the dynamics in (3). We now better clarify this point in the revised paper.
Indeed ultrasonic sensors are not very accurate, they are affected by noise and by a large field of view which causes the beam to hit different objects depending on the conditions. We do not need to explicitly account for these issues. They are well captured and managed by the kalman filter
Point 6: In semi-autonomous or autonomous systems, to achieve reliable navigation, vehicle reliable localization is desirable in any case and it cannot be connected with the season or the period of production cycle. However, authors present different distance and angle RMSE depending on ”the window” and the seasonal condition. How can authors improve this behavior? Please explain.
Response 6: The reviewer correctly points out that the localization should not depend on the season or production cycle. This is actually the main challenge that we encountered while working on the problem. The results of our research indicate that a single tuning of the EKF cannot yield a reliable and robust localization for all seasons. Either a season specific tuning is needed or, and this is the solution we propose, an adaptive module needs to be introduced. The introduction of the adaptive filtering of our solution guarantees that the EKF is robust to different seasons. Note that the window size is automatically adjusted, the localization algorithm adapts itself in order to provide consistent performance in all seasons. Please refer to fig. 14, 15, and 16 where the performance on the filter with the automatically scheduled window are more robust than the solution without the pre- selection window or even with a predetermined size of the window. We have better clarified this point in the conclusions of our work.
Author Response File: Author Response.pdf
