Generation of Realistic Cut-In Maneuvers to Support Safety Assessment of Advanced Driver Assistance Systems

Round 1

Reviewer 1 Report

This study targets a generative model based on a variational autoencoder with real-world data and then is used to generate cut-in trajectories for the safety assessment of ADASs. Overall, I think it is a well-organized and well-written paper. I just have several suggestions before its potential acceptance.

(1)  I am not sure the title is properly set because this study does not include contents related to the ADAS safety assessment,  though I am aware that safety assessment is the final purpose.

(2) This study lacks a comprehensive comparative study with other Imitation Learning-based approaches (e.g. HMM, GPR) or GAN-based frameworks for trajectory generation. Currently, it is only limited to basic math models which makes it difficult to show the unique advantages of the proposed methods. At least, supply one or two additional aforementioned methods to show their effectiveness.

(3) Supply the related parameter settings of the proposed framework to ensure its reproducibility.

Author Response

Please see the attached file. 

Author Response File: Author Response.pdf

Reviewer 2 Report

1. The paper discusses the challenges in validating and verifying advanced driver assistance systems (ADAS) for series approval, particularly in the context of level-3 ADAS like the DRIVE PILOT from Mercedes-Benz AG. It highlights the limitations of current validation approaches that use idealized maneuvers and proposes a novel approach using AI methods to generate more realistic driving scenarios. The approach involves training a generative model based on a variational autoencoder with real-world data to generate trajectories for specific driving maneuvers. The synthetic trajectories generated by the model demonstrate better replication of relevant properties of real driving data compared to mathematical models used so far.

2. The paper does not discuss the specific limitations or challenges faced during the training of the generative model based on a variational autoencoder with real-world data.

3. The paper does not provide a detailed analysis of the potential biases or limitations that may arise from using the proposed AI methods for generating realistic driving scenarios.

4. The paper does not address the potential limitations or constraints of the scenario-based validation approach using software-in-the-loop simulation, which is used in conjunction with the generative model.

5. The paper does not discuss the limitations or potential drawbacks of using idealized maneuvers as a comparison to evaluate the realism of the generated driving scenarios.

6. The paper does not provide a comprehensive analysis of the performance or accuracy of the generative model in replicating the probabilistic properties of real driving data.

7. There is no quantitative performance reported in the abstract and conclusion. In this case, it is not easy for the reader to know how good the proposed method.

8. Overall, this is a worth reading paper and as a reference that applying generative model.

9. Equation-7, the multiplication sign is too big, and not sure if it is the right symbol.

 

 

minor

Author Response

Please see the attached file. 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Thank you for addressing my concerns and I have no further questions.

Good luck with your future research.

Reviewer 2 Report

1. In 'math.' model. Do you mean mathematic model?

The authors have addressed the recommendations nicely. No further recommendation.

very minor