Hi,
A few of my solutions are feasible, but their costs can’t be calculated in the cost evaluation step on the website. I uploaded them again to see if it makes a difference but they are stuck in pending since a few hours ago. I also referred to this post and can confirm that most of these solutions have trajectories that go through lanes in the opposite direction, but this seems to be fine in other successfully evaluated solutions and the vehicle seems to be in the lanelet network at all times. You can also see in the GIF attached that the solution for USA_Peach-3_3_T-1 seems to not have any problems.
Can you provide more information on the reason for this and maybe on the method used in calculating the cost function? For example, the orientation cost in the cost function documentation is the integration of the squared difference between current orientation and the desired orientation of the state. Does that mean the desired orientation would be 0 for all time steps if the accepted orientation interval of the planning problem is (-pi,pi)? Or does desired orientation equal the orientation of a reference path calculated by the evaluation itself? If the former is true, wouldn’t that introduce an arbitrary bias towards the 0 rad orientation in the scenario, which could be any direction? The latter would also cause a bias towards the exact reference path that the evaluation prefers, which might not be optimal. Also maybe the desired orientation is the orientation of the current lane? Same ambiguity might apply to velocity and lanelet offset costs, e.g. what happens when velocity boundaries are (v1,inf) or it’s not clear which lanelet the path belongs too.
Best