find bug on [ ego-planner/src/planner/bspline_opt/src/bspline_optimizer.cpp ]

[FranklinBF]

Line 80: should be :

else if(occ && last_occ) { ++same_occ_state_times; }

[FranklinBF]

I meet some bug sometimes when running this project, and I find may be this line is wrong.

[bigsuperZZZX]

Sorry for replying so late, since I have been working on a paper submission recently. Could you describe how the bug looks like? I have exam the code but can't fig out the negative affect of the current version.

[bigsuperZZZX]

Hello, if you have no other questions, please close this issue.

[FranklinBF]

Sorry for late replying. I'm not sure if it will cause an error, just find sometimes it is not stable, and I changed it and it then has less error.

Also, the logic here, I think is to take "else if(occ && last_occ)" not just "else if(occ)“. This is what I think. I don't know if it is correct from your point of view.

I also want to thank for your work! I have use your ego planner for 2D trajectory planning in dynamic environment for my master thesis, which shows good performance due to fast replanning and not need for a collision-free initial guess.

[bigsuperZZZX]

I examed the code again. If the code is modified as you said, this line is true only if "i==order". That can cause a negative effect since the later points with "i>order" will never take as occuped in this function.

[FranklinBF]

However, if you modified as I said same_occ_state_times will still counting the points that continuously in occupany.

The following condition can be met. same_occ_state_times > ENOUGH_INTERVAL

Also, another question, why you are searching here for points with continuous same states, otherthan points with continuous occupied states?

[bigsuperZZZX]

For the modified code, since this line is always false except when "i==order_", "flag_got_start" will not get true. Then it results in the occupied segment be ignored at this line.

For the question: the reason for this trick is to make sure each segment has enough length and two adjacent segments have enough interval.

[FranklinBF]

I see. I thought it is used to find the first grid that meets a collision. Thank you very much for the detailed explanation. By the way, I meet another challenge using EGO planner in dynamic environment: If start_pos or first 3 points is in collision, then EGO planner cannot execute the optimization. Is there any solution to this problem?

[bigsuperZZZX]

It should not happen. This error means the drone is in obstacles now, which indicates a collision in the realworld. Note that we use an inflated map for planning. So please subscribe the inflated map in rviz and check why this happens.

[FranklinBF]

Thank you so much for the explanation and the issue is solved.

The problem I meant was not in the EGO-Planner project (EGO planner runs perfectly), but in my own project, where I have modified your EGO planner method for a diff-drive robot planning in an environment with crowds.

[zrd1234]

很抱歉迟到。我不确定它是否会导致错误，只是发现有时它不稳定，我改变了它，然后它的错误更少。

另外，这里的逻辑，我认为是采取“否则如果（occ &last_occ）”而不仅仅是“否则如果（occ）”。这就是我的想法。我不知道从你的角度来看是否正确。

我还要感谢你的工作！我的硕士论文中使用了您的自我规划器在动态环境中进行2D轨迹规划，由于快速重新规划并且不需要无碰撞的初始猜测，因此表现出良好的性能。 Hello, I see that you use 2D for planning, and now I also want to use 2D, may I ask how to modify the code? Thank you for your guidance