Optimizing Autonomous Wheel Loader Performance—An End-to-End Approach

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper is concisely summarized and well-written. My comments and questions are as follows.

 

C1 What is the “world model”? Please explain the definition of “world model” when the word first appears.

 

C2 In my understanding, world models are learned by machine learning. Are your proposed optimization algorithm is effective for unknown world models? You should add several passages to answer why this is effective or not to unknown world models.

 

C3 Is this algorithm online or offline? Please mention this in the manuscript.

 

C4 Why greedy method is most effective in Table2? Do you have any theoretical insight about this. Please mention insight in the manuscript.

 

C5 Please add nomenclature.

 

C6 Please explain the influence of initial conditions and random seeds on the optimal strategy obtained. Is strategy stable or unstable?

 

C7 You should summarize the values of parameters used for simulation in table form.

 

C8 What value of weight w is used for simulation? If you used one value of w for simulation, please conduct simulations under different types of w. The different results might be obtained. So please discuss this in the manuscript.

 

C9 Why do you not conduct simulation in different soil parameters. In my understanding, soil parameters can be changed numerically and you do not need to take field data.

 

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript entitled “Optimizing autonomous wheel loader performance - an end-to-end approach” presents a data-driven framework for optimizing sequential loading operations in automated wheel loaders. The proposed look-ahead planning method, together with the integration of world models and low-level control models, is clearly described and supported by experimental simulations.

 

While the paper is overall well-written and technically solid, there are several aspects that would benefit from clarification or elaboration. Addressing the following comments could improve the clarity, completeness, and practical relevance of the work:

 

1. The related work section could benefit from more structured comparisons. For example, organizing prior studies based on whether they focus on pile dynamics, energy efficiency, or planning horizon would help clarify the novelty of the proposed approach.

 

2. The multi-objective performance function is weighted by a vector w, but the rationale behind the specific choice of weights is not explained. Please briefly state whether these values were tuned, fixed heuristically, or derived through optimization.

 

3. As the method relies heavily on the accuracy of the predictive world model, it would be helpful to include a brief summary of its training process-such as dataset size, input representation, and validation setup-to enhance credibility and reproducibility.

 

4. The use of B-spline curves for V-turn path generation is appropriate. However, a brief explanation or a sentence comparing it with alternatives like Dubins paths would provide more context for readers.

 

5. While per-step computation time is reported, the paper does not elaborate on how action candidates are managed during tree search. Please clarify whether any pruning, sampling, or heuristic techniques are employed to reduce complexity at higher depths.

 

6. The formulation assumes flat terrain, fixed receiver location, homogeneous soil, and no spillage. These are reasonable simplifications, but a brief acknowledgment of how they may affect real-world applicability would enhance the discussion section.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I think the authors respond to my questions and comments properly. So, the manuscript can be published.

Author Response

Thank you again