Design and Implementation of a Real Time Control System for a 2DOF Robot Based on Recurrent High Order Neural Network Using a Hardware in the Loop Architecture

Round 1

Reviewer 1 Report

1. Please, prove that this type of research or this type of controller design has a value for the research community. In your reference list there are only 2 papers (10%) that are younger than 5 years old, that can show rather weak interest to the topic in modern literature. It would help a lot to landscape the state of art in solving this type of problem before you suggest your solution.

2. Please, provide limitations of your results. For example, your experiments demonstrate an excellent performance of your way to identify and control the robot, but can you guarantee that the algorithm is robust in the case of 2DOF robot with different physical parameters or unmodelled dynamics?

3. Could it be so that the references signals (24) and what is shown as reference signals in Fig. 6 a) and b) are different?  At least the magnitude of x1 is to be 1.5 according to (24) but is is obviously 1 in the Fig 6 (a). It is also not clear why the frequency in (24) are chosen to be varying with time "ω1 = (3.5 t + 1) rad/s and ω2 = (3 t + 3) rad/s" and if they are chose this way why in Fig. 6 we see constant frequency of the reference signal.

4. The paper is written in very clear and understandable language. Nevertheless, an extra proofreading is always good. Examples:

string 176 "shown on Figs. 5a) y 5b) explained" - "y" should have been "and", most probably,
string 179 ". These reference signals have been chosen arbitrarily due to demonstrate" - not sure it is legitimate grammatically,
the style of variables in the text, formulae and figures should be the same,
string 219 "The results of tracking control and Identification process shows the high performance" - Capital in "identification" is not needed, "shows" should be "show" as plural. 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

This manuscript proposes a real-time implementation of a SMC to control robots, concretely the algorithm is tested in a simulated 2DoF robot. The authors propose the use of RHONN with a modified Extended Kalman Filter to identify the robot model. The proposal is sustained experimentally, using a 2DoF model in Simulink and the controller implemented in a Virtex 7 FPGA.

The paper is well written and easy to follow. Although there is a lack of theoretical contribution, the authors present an interesting real time implementation of SMC control.

I have some concerns that should be clarify before thinking in its publication:

• A 2 DoF robot model is a very simple model in order to generalize an approach. Since the use of a real robot to validate completely the approach could be problematic at this stage, at least the authors should consider to simulate with a 6DoF robot, which is more general.
• The authors should give more details about nonlinearities present of the modeled robot (i.e., motors death zones, stiffness, etc.). This will justify the use of Neural Networks to model the robot, otherwise classic identification model algorithms could be used.
• Covariance matrices of the EKF are computed using the dynamic model and the authors claimed that this is a key contribution of this work. However, it can be found in the literature lots of approaches using dynamic models for the EKF. Authors should soften this claim or specify the difference between their contribution and the ones in the literature.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The revised manuscript has been improved according to the reviewers' comments and can be published in its current state.