navfn (what we currently have) is outdated, and supposed to be replaced by the more flexible global_planner
carrot_planner is too simple and literally just draws a straight line to the goal point and does not do any obstacle avoidance - not an option for us
D > A for unknown/very dynamic maps for online replanning (which is what we have on a campus) - can't find an easy to use implementation for this right now, but this can be a recommendation for the future
Global Planner Configuration
A* algo > Djikstra as it has a heuristic for distance and is faster at searching
Quadratic approximation of the potential > simple calc (unless we are running into issues with processigng)
Gradient descent > grid search bc : gradient descend is a disciplined mathematical operation, which is guaranteed to find the global minimum for convex functions (although NN loss functions are not convex), while grid search is actually just an ad hoc, quick-and-dirty procedure which does not guarantee anything , BUUT A* ROS implementation doesnt work properly with gradient descent, so we will use grid search
orientation_mode = forwardtheninterpolate - this is because our ultrasonic sensors are in the front, so we should try to always plan to move w the robot facing forward
allow_unknown: true - allow the robot to traverse unknown space so we can plan through this area
navfn(what we currently have) is outdated, and supposed to be replaced by the more flexibleglobal_plannercarrot_planneris too simple and literally just draws a straight line to the goal point and does not do any obstacle avoidance - not an option for usGlobal Planner Configuration