About default feature range in Kinematic Observation

Hi!

I’m curious where the default features range in KinematicObservation come from? To the point I know, the x value can be from 0 to maximum_duration * MAX_SPEED, but in that line, it is considered in (-5 * MAX_SPEED, 5 * MAX_SPEED) range. So I can’t understand:

• Why is the negative range also considered for x and y?
• Why is maximum duration not considered?
• Why five times of MAX_SPEED?

Again I don’t understand why the speed range is two times the MAX_SPEED.

Thank you

[eleurent]

Could you mention a code line in which you’ve changed the origin? I wanted to see how it is possible.

You cannot really change the origin, but rather the position of the road network (relative to the origin). In intersection-v0 or roundabout-v0 for instance, the origin is in the middle of the intersection/roundbout.

but isn’t the point of normalization that we can always expect the normalized value won’t exceed 1?

Well, not necessarily. It depends on why you are normalizing, really. If it is related to model conditioning for function approximation, it doesn’t have to be a hard constraint, and people often normalize by dividing by the standard deviation (which means you can still get a value >= 2 with ~13% probability). See e.g. https://www.jeremyjordan.me/batch-normalization/, https://towardsdatascience.com/why-data-should-be-normalized-before-training-a-neural-network-c626b7f66c7d

I think it is up to you to choose the proper feature range / scale depending on your particular needs.

[eleurent]

Well it can be both, depending on whether the field absolute is set to true or false. But even in the case of absolute coordinates, in some environments the frame origin is located such that vehicles can have negative coordinates. The horizontal roads of highway-v0 are a special case.