You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, the ball search heatmap is very basic, since it decays all cells independent of each other and the state of the Nao. An improvement would be to cooldown cells in the viewing cone of the Nao up to a certain distance faster than others. This would also fix our Naos from running into one corner during corner kick and not checking the other one (related to #1288)
The text was updated successfully, but these errors were encountered:
I'm afraid just decaying the values where we don't see a ball won't help in the corner kick case.
The current search suggestor simply overwrites the heatmap values in both corners with a static 1.0 every cycle during the corner kick.
The correct way would be to only set these values when entering the corner kick state.
I'm still convinced the ballsearch heatmap is barely functional in its current state, see #1668 for more ideas.
Idea: reproject field segments onto the ground and decay the heatmap accordingly:
We can assume that the ball is not in a location determined to have field color.
Edit:
Better yet, project the heatmap into the image, this makes everything less complicated since segments are now always contiguous and non-overlapping. I have partially implemented this, but am unsure if this is really what we want. More discussion is warranted
Currently, the ball search heatmap is very basic, since it decays all cells independent of each other and the state of the Nao. An improvement would be to cooldown cells in the viewing cone of the Nao up to a certain distance faster than others. This would also fix our Naos from running into one corner during corner kick and not checking the other one (related to #1288)
The text was updated successfully, but these errors were encountered: