Open namdre opened 3 years ago
The problem was introduced via #796 (version 0.16.0) nine years ago!
The problem is likely affected by the accuracy of the estimated times which varies by intgration method (--step-method.ballistic). The initial test shows one collision per 13k left turns with ballistic integration and 0 collisions within 90k left turns for default euler integration.
Right now, the lack of post-encroachment gap somewhat compensates for the discrepancy between conflict area and junction area when estimating conflict times. Thus, a fix here makes it more important to fix other things as well (#1846)
a new junction model parameter shall be introduced to configure the follow-from-minor-timegap independent from lead-from-minor-timegap
If the (estimated) ego arrival time at the link comes later (even if by a millisecond) than the estimated exit time of the foe and the foe has a different target lane, the link is deemed free.
In contrast, a time gap is enforced when passing ahead of a foe (default 1s, configurable with jmTimeGapMinor). The same time gap is also enforced if the foe is leading and the ego vehicle targets the same lane.
Usually, this isn't a problem since the estimated arrival and exit times are with reference to the whole junction shape which gives a sufficient safety buffer. When passing via an internal junction, the vehicle is waiting (almost) directly besides the conflict area and the lack of buffer may cause collisions or low post-encroachment-time.