Hi,
I started on working on the heuristic as suggested in the pdf. I am currently using the tutorial_commonroad_search for testing my heuristic on the given example and the tutorial_batch_processing for testing on the whole 500 scenarios. As this parallel bach processing takes a long time, if I would run it every time for all changes on my heuristic, I wondered if there is an alternative, some kind of best practise, to test the heuristic beforehand on a smaller subset, to be able to tell if the heuristic improved or not without having to run all 500 samples every time.
And one second thing: I am working on a heuristic for the Astar algorithm, which means I have to calculate the path cost, which is added to the heuristic in the evaluation function. The heuristic always has to underestimate the cost to reach the goal (my understanding: final path cost > calculated heuristic), and for example for the distance this makes perfect sense, but for example for the difference in orientation, I am not sure how I would define the path cost for prior orientation differences. And also how can I ensure this behaviour when I introduce different weights. Do I introduce weights in the path cost as well? Or is this weighted heuristic approach in general more applicable to greedy best first search? Or maybe I have a missunderstanding here.
Answers on those questions would be highly appreciated.
Best regards
Kai