Online Motion Planning for Safe Human–Robot Cooperation Using B-Splines and Hidden Markov Models

Round 1

Reviewer 1 Report

The paper presents an approach for online motion planning for safe human-robot cooperation using B-Splines and Hidden Markov Models. The paper is overall interesting and easy to read. The proposed approach is supported by experimental results. However, the following points need to be clarified to improve the overall quality of the manuscript.

·       The main novelties of the paper should be better highlighted with respect to the present literature. The authors should better describe the advantages and disadvantages of the proposed approach with respect to similar work on the same topic.

·       It is not clear which kinematics and dynamics constraints are considered in the implementation of the proposed approach on the Franka arm.

·       Values of mi, bi and ki are positive constants such that the system is critically damped. The authors should report the values of the constants and how those constants have been derived.

·       It would be suitable to include the plots of the trajectory of the robot (e.g., joint variables over time, repulsive virtual force over time, human-robot distance over time).

·       What is the accuracy of the detection of the marker positions on the hand of the human operator by the tracking system?

·       What is the rate of the Cartesian controller implemented in c++?

·       Section 4.1 should be organized better, and the different experiments performed on the robot should be better explained and detailed.

·       Line 295, definition of “success rate”: it is not clear how the safety threshold has been tuned/set and how the threshold influence the success rate.

·       Line 313: a point is missing.

·       How is the robot steered to stop? How does it restart its motion after a safety stop?

·       The quality of the figures should be improved. In Figure 9(a), is the computational time expressed in ms? The 10^-3 should be removed.

·       Values in Tab.1 should be reported as mean +/- standard deviation. Why there is no value for Tstop in the B-Spline case?

·       “On one hand” should be used in combination with “on the other hand”.

·       The literature should be improved, providing examples of “many other publications of the topic” as stated in line 82 in Section 1.1, especially related to the speed and separation monitoring mode of interaction. Some suggested references are the following:

o   Enhancing fluency and productivity in human-robot collaboration through online scaling of dynamic safety zones. The International Journal of Advanced Manufacturing Technology, 121(9-10), 6783-6798. 2022.

o   Maximising Coefficiency of Human-Robot Handovers through Reinforcement Learning. IEEE Robotics and Automation Letters. 2023.

o   A collision-free path planning method for industrial robot manipulators considering safe human–robot interaction. Intelligent Service Robotics, 1-37. 2023.

o   Real-time motion control of robotic manipulators for safe human–robot coexistence. Robotics and Computer-Integrated Manufacturing, 73, 102223. 2022.

Author Response

The review was updated as a pdf file where it is possible to find:

• point-by-point response to the reviewer's comments;
• the modified article with modifications written in red.

Please note that, as we applied the same strategy to the other review, the final manuscript will be comprehensive of every modification required by each Reviewer.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper presents a solution to dynamic obstacle avoidance by a robotic manipulator. The proposed solution uses two components: modification of trajectory control points determining B-spline trajectory and modification of velocity to slow down or stop the motion before a collision which uses Hidden Markov Model. The main contribution of the author is combining the two approaches in a single method with a parameterized influence of both components.

The paper clearly describes the problem and presents the authors' proposition of its solution. It would however improve the quality of the paper if the authors tried to formally show the validity of the method, not relying only on simulations. It also seems that the method was designed with optimistic assumptions, that were not clearly stated or discussed. Specifically I would like the authors to discuss:

- if and how robot dynamics is included (and if not - what are the assumptions on latency and velocity/acceleration limits)

- how repulsing a control point influences the trajectory (even a big shift of a control point may not move trajectory enough for it to be non-collision);

- the guarantees that the algorithm of velocity control described in l. 212-222 allows indeed to stop before the collision occurs

- how the algorithm behaves if the obstacle cannot be avoided either be trajectory shift or velocity decrease

In the presentation of the test results the success rate is below 100% what means that the distance drops below the assumed safety gap. It would be good if details were presented, for example complementing experiment plots showing  robot to obstacle distance in time and providing the value of minimal distance observed. I would also wished to see  more structured tests, focusing on various aspects and presenting those scenarios - currently the reader must rely on one experiment  with obstacle motion seeming random.

Finally there is a disturbing issue with references - after checking first 6 of references, 5 of them had DOI that was "not found" or leading to a different work than cited. I wish the authors checked all the cited works and verified that the data they provide are correct.

Author Response

The review was updated as a pdf file where it is possible to find:

• point-by-point response to the reviewer's comments;
• the modified article with modifications written in blue.

Please note that, as we applied the same strategy to the other review, the final manuscript will be comprehensive of every modification required by each Reviewer.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Thank you for the responses. The paper has been improved with respect to the previous version. 

Reviewer 2 Report

The updated version of the paper clarifies most issues raised in the previous review. One could wish for more test cases in various scenarios confirming the approach, but I understand that it would require additional time and it would make the paper much longer.

There was a small detail that I would suggest authors to look at when preparing the final version:  "accuracy lower than 4cm" (l.397) seems unclear to me. While I assume that it means cases when Cartesian error in 3D is greater than 4cm, maybe the authors could reword it.