Reviewer Comments and Actionable Points

Reviewer Comments and Actionable Points

Reviewer #1

•
Time Complexity Discussion:
- –
  
  Action: Discuss the computational (time) complexity of the proposed method, particularly for both training and inference phases.
- –
  
  Context: Real-world control systems often have strict latency requirements, so it is important to explain how the proposed method handles this, or any performance trade-offs.

Reviewer #2

•
Comparison with Traditional Methods:
- –
  
  Action: Include a comparison with traditional methods, such as numerical solutions of the Fossen model, in terms of accuracy and efficiency.
- –
  
  Response: We have already done this comparison in the final expriments which is ground truth (simulation model) comparison.
•
Control Task Performance:
- –
  
  Action: Clarify whether the neural network model is used for control tasks, as the name PINC suggests. Discuss the model’s performance on real-world data or plan future work to explore this aspect.
- –
  
  Response: It is clear that this work focuses on control, as discussed throughout the text. For example, ”This work explores the potential of a specialized variant of PINNs, namely the physics-informed neural network with control (PINC) [6], for modeling the dynamics of underwater ROVs. The primary objective is to evaluate whether PINC can effectively model a simplified underwater dynamic system and provide an accurate model for control applications.” For future work, have added the following: Training PINC models on real-world ROV trajectories would enable a more realistic evaluation of their efficacy, potentially uncovering limitations not captured in simulation.

Reviewer #3

•
Novelty of the Work:
- –
  
  Action: Discuss the novelty of the work in comparison to existing literature, particularly PINC applications like Section 4 of [6].
- –
  
  Context: Although PINC is well-established, the application to underwater robotics and your specific system may present new insights that should be highlighted.
•
Comparison with Traditional Methods:
- –
  
  Action: Similar to Reviewer #2, include comparisons with traditional methods (e.g., numerical solutions of the Fossen model) for accuracy and efficiency.
- –
  
  Context: This will help demonstrate the relative effectiveness of the proposed method.
•
Clarification of Loss Functions:
- –
  
  Action: Provide a clearer explanation of the loss functions, particularly the $f$ term in the physics loss (equation 2) and the physics rollout loss (equation 5).
- –
  
  Context: A clearer explanation of these terms will help reviewers understand the mechanics behind your method.
•
Clarification of Development Set:
- –
  
  Action: Clarify the process of increasing the development set size to match the test set. Specifically, explain whether the model ”saw” the part of the development set during training.
- –
  
  Response: Added Remark 4.