Fix coordinate transforms

[cbachhuber]

Continuing #23: Replaced all unknown transforms from the simulation and evaluation chain except angle regression by well-understood transforms.

• Fix coordinate transforms in simulator and precision evaluator.
  • In the measurement grid, vehicles are now at the coordinates defined in their constructor. Examples: Vehicle at position (10,20) will be at pixels (10/resolution, 20/resolution). A vehicle with position (grid_size, grid_size) will be at the top right of the grid.
  • Same for velocity: I verified that e.g. a vehicle with a speed of 10m/s covers 10/resolution pixels in one second.
  • Both simulator and evaluator now take grid size, grid resolution, number of scan points, and sensor FoV into account. They work for arbitrary values (within a sensible range 😉 ) in these four variables.
• Vehicle positions now denote their center. Makes precision evaluation simpler
• In code, I noted that laser.max_range must be grid_size, otherwise vehicle positions are incorrect. Same for resolutions.
• Removed visual clutter from demo output

I think this is a substantial improvement for simulation and evaluation. Now we understand them almost entirely.

[TheCodez]

Thank you 👍 Cool to see improvements here!

In code, I noted that laser.max_range must be grid_size, otherwise vehicle positions are incorrect. Same for resolutions.

I just pushed a commit to fix this, but I want to further change this (the line above my change).
Currently the ego vehicle is at the bottom of the map. In the future I want to move the vehicle to the center of the map as in the paper (in the paper the images are actually cropped so it looks like our current implementation with the vehicle in the bottom).

[cbachhuber]

Ok cool, I rebased on your commit. With that, only the resolution dependency remains 👍