Comments to author (Associate Editor) ===================================== The authors proposed a method to learn personalized driving styles for drivers with limited data. Besides comments from individual reviewers, major concerns are listed as below: ï Environment impact Vs Driver's own driving styles, it is not very clear how to differentiate the driver's action is related to the driving styles or due to some other environment changes, Some explanation can be added for clarification. ï ACC systems variations: Some details can be added to describe the differences on ACC systems involved. And will that affect driver's decision? Reviewer 6 of ITSC 2022 submission 428 Comments to the author ====================== The paper proposes a method for learning personalized adaptive cruise control models based on observations of a particular driver's behavior on a small number of instances of following a leading vehicle. The driver's preferences are discovered through inverse reinforcement learning in a discretized POMDP, which considers the speed and distance to leading vehicle as observations and acceleration as the action. Overall, the paper goes into great detail explaining what the authors did, but it doesn't explain why they made those particular choices and how the results compare to state of the art. I don't quite understand the evaluation metrics used, but it seems that reported accuracy is evaluated with respect to the outputs of some other custom model rather than real observations. There are no baselines so I have no way of judging how this method performs compared to existing alternatives. Unfortunately, I'm not familiar with the literature in this area, so I can not suggest suitable baselines here. The contributions stated in Section I are phrased in terms of what methods the authors used, rather than what performance they accomplished or what hypotheses they validated. The paper reads like a system description lacking proper evaluation or statements of concrete findings. Other comments and questions: 1. The authors elect to use a model-free IRL method, indicating that model-based approaches require knowing the transition dynamics, but it seems to me that those are available in this simplified driving simulation. 2. Related work lists existing papers in the area, but it doesn't provide a synthesis that would allow the reader to see the big picture. 3. It seems that the setting is very simplified, in particular regarding discretization into a small number of bins. This is not necessarily bad, but I would appreciate some discussion about how far the system is form being practical. 4. The writing could use some improvement. There are a lot of sentences I couldn't make sense of, such as "In order to evaluate that which cluster of drivers that the driver in the target set is most similar to, we compared the KL-divergence". Reviewer 9 of ITSC 2022 submission 428 Comments to the author ====================== This paper develops a method to learn a driver's personal driving style from limited data for personalized adaptive cruise control. The method is based on combination of inverse reinforcement learning and other techniques. I think it is appropriate for the conference. The state space is discrete with large discretization widths for the driving phenomenon. For example, vehicle-to-vehicle distances from 24 m to 48 m are indistinguishable. Is this safe? What will happen if the discretization width are decreased? Reviewer 11 of ITSC 2022 submission 428 Comments to the author ====================== The paper is concerned with the method to learn and simulate individual driving styles so that the driver's preferences can be appropriately taken into account. The method is interesting and the simulation results are convincing. Therefore I recommend the paper for publication, after correcting some small typos, e.g. "can be learn" --> "can be learned" in Introduction. A small question to the authors: In Conclusion you wrote "adapt the personalized model to a more complex driving situaion". Could you explain about it a bit more? (It does not need to be included in the paper itself.) Is it also possible to consider how to avoid an obstacle? With an obstacke we can also observe many different reactions/preferences by drivers. Reviewer 13 of ITSC 2022 submission 428 Comments to the author ====================== This paper addresses a method to adapt ACC parameters based on driving style of a driver. They combined clustering method, IRL-based reward learning, and POMDP solver to adapt the parameters to the driver. They used dataset and evaluated the performance using the drivers which have enough number of events. While I believe this is a solid result, I have several concerns. Major points: >The model-based experimental driving trajectories demonstrate that the P-ACC system can > provide a personalized driving experience. I think this claim is not supported enough by the results. 1. Gap between how drivers would like to drive and how they would like to be driven [1,2] ñ Studies suggested there is a gap between how they drive and how they prefer to be driven by automation. E.g., [1] indicated that 30% preferred opposite driving style to theirs. 2. Study still not complete: They demonstrate the user can be adapted to (clustering-based) types. However, it does not mean adaptation to individual is achieved. User study will help understand the effectiveness of the approach. 3. Very simple parameterization for ACC ñ as the authors describe in the conclusions section, only speed and following distance. ACC should include maximum acceleration/deacceleration and behavior when cut-in events happened. Overall, solid study but could be improved with more follow-up studies. [1] Stefan Griesche Bosch, Mandy Dotzauer, Stefan Griesche, Eric Nicolay, Dirk Assmann, and David K‰thner. 2016. Should my car drive as I do? What kind of driving style do drivers prefer for the design of automated driving functions? Should my car drive as I do? What kind of driving style do drivers prefer for the design of automated driving functions? Soll mein Aut. raunschweiger Symposium 10, 11 (2016), 185ñ204. [2] Franziska Hartwich, Matthias Beggiato, and Josef F Krems. 2018. Driving Comfort , Enjoyment , and Acceptance of Automated Driving - Effects of Drivers í Age and Driving Style Familiarity. Ergonomics 0139 (2018), 0ñ1.