Currently, for the prediction task, the get_closest_lane method can get the closest lane to the given point. But this closeness is defined in norm distance manner. But sometimes the car is a little bit off of the lane, in this case, if we still use the get_closest_lane method, we may get the lane that is physically closest to us but in the wrong direction. For example, if the car at the stop area at intersection and the car is about to turn right. But the closest lane can be the left lane since the annotated turn right lane at the intersection is off from the current position. So would it be possible to also consider the direction of lanes when defining the closeness?
Closest lane definition
Hi. While I think this is a legitimate use-case, I think it would change the method too much, as you’d also need to provide the ego vehicle orientation. I suggest you write your own method which takes these additional inputs and applies the direction constraint.