Algorithm for Base Action Set Generation Focusing on Undiscovered Sensor Values
Round 1
Reviewer 1 Report
Overall its a good work, but quality can be improved
Improve on language and shorten the sentences where needed
The paper compares Algorithm Random vs Method A. Is it possible to compare the results with actual base line or ground truth values and provide error rate of the proposed methodology
Can you please elaborate more in the theory, how S(t) is obtained from bin values from prior sensor values i.e. from s(t-1)
Author Response
Response to Reviewer 1 Comments
Point 1:
Improve on language and shorten the sentences where needed
The paper compares Algorithm Random vs Method A. Is it possible to compare the results with actual base line or ground truth values and provide error rate of the proposed methodology
Response 1: Thanks for your helpful comments. It is not possible to get actual base line in this case because the actual number of box patterns that the robot can discover with ideal locomotion during experiment time is unknown. This is why we prepared Algorithm Random and compared Algorithm A with it.
Point 2: Can you please elaborate more in the theory, how S(t) is obtained from bin values from prior sensor values i.e. from s(t-1)
Response 2: Sensor values s(t) and s(t-1) are independent and histogram Dm and its bins are used to determine whether s(t) is an undiscovered sensor value or not. We have added further explanation of sensor value s(t) to Section 3.1.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors have done a major review of their previous version of the paper. The paper presented an action set generation approach. The problem that I have with this work is it does not compare its results against a similar approach. Also, I believe that the sensor input to sensor output learning has been studied before and authors should address how this work is different from the previous work [1].
[1] Dutta, A., Dasgupta, P., & Nelson, C. (2017), Adaptive locomotion learning in modular self-reconfigurable robots: a game theoretic approach, In IROS 2017 (pp. 3556-3561), IEEE.
Author Response
Response to Reviewer 2 Comments
Point 1:
The authors have done a major review of their previous version of the paper. The paper presented an action set generation approach. The problem that I have with this work is it does not compare its results against a similar approach. Also, I believe that the sensor input to sensor output learning has been studied before and authors should address how this work is different from the previous work [1].
[1] Dutta, A., Dasgupta, P., & Nelson, C. (2017), Adaptive locomotion learning in modular self-reconfigurable robots: a game theoretic approach, In IROS 2017 (pp. 3556-3561), IEEE.
Response 1: Thanks for your helpful comments. Previous studies including [1] proposed methods to acquire actions for the specific tasks. These methods prepared a reward function and used reinforcement learning methods for achieving the tasks. The purpose of the proposed algorithm is to acquire a universal action set which does not depend on a specific task and the aims of this study is different. Thus, it is not possible to compare the results of previous studies and results in this paper. We have added further explanation about [1] to Section 1.
Author Response File: Author Response.pdf
