GTU - Lane bookkeeping

[WJSchakel]

It is unlikely that OTS will function sufficiently fast in all cases if it is designed in a complete space-based manner. That is, if objects can find each other through spatially looking each other up. OTS should provide interfaces for information that perception will use, for which different underlying approaches may be implemented. In particular it likely makes sense to keep a registering of GTUs on lanes, but only by their reference point. Initial ideas are:

• GTUs are registered to 0 or 1 Lane at any time.
• When GTUs enter (their reference point) another lane, either laterally or longitudinally, they register on the next lane and unregister on the previous lane.
• Registration based on reference point should not give the same bugs as registration with front/rear. There are inherent problems (bugs) in terms of jumps and discrepancy between extrapolated and actual path, being a combination of a straight line to the front, and a curved path of the reference point.
• GTUs remain registered to their previous lane in case of small gaps between lanes.
• If a GTU wanders away from a lane without entering another lane, it should unregister based on some distance heuristic.
• GTUs can be found by perception, even if e.g. only their rear is on a particular lane, as the reference point will be on another lane. Perception needs to look towards several adjacent lanes, and downstream over some viewing distance, anyway.
• Perception can use the lateral position of GTUs to assign them on lanes, perhaps even assigning them to multiple lanes.
• There is currently no easy method to obtain the lateral position.
• These two methods are likely handy to have for LaneBasedGtu: Lane getLane() and Length getLateralPosition().

Before going through with this, we need to consider relevant use-cases.

[WJSchakel]

In the new system, Detectors will work autonomously. They may use the GTU-lane bookkeeping, or be space-based. Various approaches of different accuracy and efficiency can be used. For example only being triggered by the reference point, or using a simple length-based heuristic. But this is very detector specific, and hence better placed purely at the detector side.