• the obstacle tracking / matching algorithm frequently double counts or treats existing obstacles as new ones.

This is an example of where i believe it is going wrong and one way to fix it, but I think it can be improved too.

• currently, obstacle tracking matches global positions of obstacles btwn time frames to determine that they are the same obstacle. however, the global position is inaccurate, especially when the robot is turning (or, more generally, when the obstacles relative positions to the robot are changing quickly)
• modify obstacle tracking so that it ignores (or in some other way treats differently / takes note of) measurements of obstacle locations that are taken when the robot is turning
• "when the robot is turning" is to be better defined. ie. should it be some threshold of the robot's angular velocity? or a diff btwn this and last frame's relative obstacle positions? not really sure.