[Feature]: Lanelet2 to BEV Map rendering update

[m-zain-khawaja]

Problem / motivation

We need to encode route information in our map input to the network when we are relying upon a Lanelet2 map.

Proposed solution

Rendering

• We should render the Lanelet2 map in a 256px x 256px square grid of resolution 0.5m
• We can take the ground truth waypoints from the vehicles motion and create a one-hot (black/white) image encoding the route data of the same size and resolution as the map render
• Both of these inputs can be fed into the network to train
• The route should be encoded up to the edge of the map view
• If it is possible, we can try snapping the waypoints to the centre-lines taking car at lane changes, where for example, if the driven path doesn't snap neatly then we do some interpolation.

Inference

• Prior to an actual autonomous driving run, a user would have to annotate the route they want the vehicle to take offline and we would use that to create the one-hot encoding of the route

Alternatives considered

No response

[m-zain-khawaja]

@bharatwrrr - could you please take this on

[bharatwrrr]

@m-zain-khawaja, I think #161 closes this issue as well.