search

ZJU-FAST-Lab / LBFGS-Lite

LBFGS-Lite: A header-only L-BFGS unconstrained optimizer.
MIT License
321 stars 57 forks source link

Maximum feasible cost allowed after a optimisation step? #9

Open shubham-shahh opened 1 year ago

shubham-shahh commented 1 year ago

Hi, thanks for this awesome work @ZhepeiWang I am using ego-swarm-planner-v2 which uses lbfgs.hpp, I am not sure if it's related to this repo, but your help is much appreciated.

recently, when I was trying ego-swarm-planner-v2, I observed, the drone made a path with cost = 1794351201.188. and the generated path was not good and the drone started executing the path, I am curious, what's the maximum "feasible" cost I can allow?

below I have attached the image with cost and path

thanks and regards imageedit_2_5660280720

shubham-shahh commented 1 year ago

Hi @bigsuperZZZX, your help is much appreciated

bigsuperZZZX commented 1 year ago

The maximum "feasible" cost depends on your implementation. From an engineering point, you could record the normal cost when planning in the most cluttered environment possible, and then the maximum "feasible" cost can be set several times larger than the normal cost just recorded.

shubham-shahh commented 1 year ago

The maximum "feasible" cost depends on your implementation. From an engineering point, you could record the normal cost when planning in the most cluttered environment possible, and then the maximum "feasible" cost can be set several times larger than the normal cost just recorded.

I see, that makes sense. If I try to introduce a check with a max cost, I am facing issues, as mentioned here https://github.com/ZJU-FAST-Lab/EGO-Planner-v2/issues/12#issuecomment-1410228620