2) Models are known to behave differently with different step-lengths
- this is partly due to the effects of extreme accelerations caused by euler integration so you can reduce the effect by using ballistic integration. This should also be noticeable at intersections
- another source for differing behavior is the effective reaction time which is currently tied to the step-length (vehicles can choose to change their speed during every step). We are working on an extension to decouple the decision making interval from the step length but it's not quite ready yet.
- there are some known issues with ballistic update where vehicles take longer to find a gap for lane-changing. The vehicles are somewhat conservative with their safety constraints which is less noticeable with the higher accelerations afforded by the euler integration. The effect should be less pronounced with lower step-length.
- generally, we intend to make ballistic update the default behavior because it generates more realistic trajectories. If you notice something odd your are likely to get support fairly quickly so I would already recommend to use ballistic integration.
regards,
Jakob