[Done] Explain better that the probability distribution map is not a mixture of Gaussian to start with. It's more complex. And we also need to combine it with the task difficulty map. We are just using GMM to approximate.
[Done] Distinguish presented work from previous work (from IROS paper)
[Done] Our paper is not using a Bayesian Approach, but a combinatorial optimization problem.
[Done] What are the factors that motivate choosing GMM?
[Done] Make it clear that we are multiply two things together assuming independence and they might not be independent.
How does Mode Goodness Ratio compare to other metrics?
[Done] Don't use arbitrary number of Gaussian. Need to explain that when computation time is critical, then even arbitrary number of Gaussian is better than Bourgault's algorithm and LHC-GW-CONV algorithm. When there's a little bit of time, then hierarchical search produces much better paths.
[Done] Make it clear that the previous LHC-GW-CONV algorithm works well when task difficulty is low and also handles multi-modal distributions well.
[Done] Make it clear that the new algorithms are very different from previous LHC-GW-CONV algorithm, so performance increase has nothing to do with building on top of previous algorithm.
[Done] Mention Vehicle routing problem in related work (together with TSP). Repeated visit adds more complexity.
[Done] Do a better job explaining what LHC-GW-CONV means and what the algorithm does.
[Done] Discuss advantages and disadvantages of each algorithm.
[Done] Explain task difficulty map. Can be treated as sensor limitation (e.g., vegetation density)
[Done] Explain how are the GMMs learned (using first K-means, then EM to iteratively learn GMMs)
[Done] Figure 4 is not clear. Need more graphs and text to illustrate the algorithms better
[Done] Explain how are modes identified?
[Done] Explain how we sample from distribution map in order to do K-means and EM.
[Done] What's the sensitivity of number of Gaussians? (really depends on the situation)
[Done] Discuss false alarms, moving targets, changing grid size, etc.
What's the time spent on GMM vs. the rest?
[Done] Make it clear that the teleport path is not optimal because it is more than optimal. And our algorithms approximate optimal paths.
[Done] Do something about the final paths figures because they are difficult to see in print.
[Done] Don't use the word “agent” because it's confusing.