New lane structure

[WJSchakel]

The current lane structure assumes one record at the longitudinal coordinate of 0m, on each relative lane. E.g. that there would be one record for one lane to the right of the GTU. However, this is not true. For example:

• Two acceleration lanes (at merges) can be close together. The lanes leading up the acceleration lanes are both to the right of a GTU on the right-hand lane of the mainline.
• If shoulders are included to also possibly have vehicles, for example as they were unable to merge in time, the shoulder too must be considered to the right.

Therefore the lane structure must supply a set of lanes, from which neighboring GTUs and objects can be found. Many perception categories and classes such as lane change incentives rely on taking a single record to the left or right, to derive information in a search starting from this record. These algorithms all need to be updated.

[WJSchakel]

The new LaneStructure builds the structure every time step, but only to a limited range of about 300m downstream and upstream. Longitudinal search has absolute priority over lateral search. Lateral search will skip any lane that was already found longitudinally.

In any upstream search, when a lateral move is made, the relative lane of this laterally found lane remains the same as the lane it is found from. This is because if a lateral move is made, it is the only way the GTU can ever get in interaction with the ego GTU. For example: when the left lane merges with the right lane before reaching an acceleration lane, traffic on this left lane of the two can be laterally considered to be on the same lane as the acceleration lane.

The class can provide:

• Leading and following GTU's on all lanes in the structure, sorted near to far.
• First leading and following GTU's on all lanes in the structure, sorted near to far. This only contains the first leader, or if there is a split before any GTU, the first leader on each branch. This may repeat, and also applies for followers.
• Upstream and downstream LaneBasedObjects of a given type, sorted near to far.

[WJSchakel]

• The rolling lane structure has been removed.
• Perception categories have been extensively updated to use the new lane structure.
• Incentives have been updated to use the new structure. IncentiveRoute uses the shortest path lane change info algorithm from RoadNetwork to go beyond the extent of the lane structure.
• Conflicts still use a simple version of lane records, within perception iterables.
• Lane records within the lane structure are also simpler, as they are no longer laterally connected. This was only used for lane change info, which is now taken from the network.

Overall the data structure is simpler and more robust. GTU's can now use the shoulder if they need to.

[WJSchakel]

A small test with the short merge demo revealed that where the old lane structure took a total of 3662ms to be created during one hour of simulation, this was only 26ms for the new data structure.