Egovehicle Crashes and Priority Stopping Mechanisms in Autoware

[ExitedState]

Issue Description
From the three videos below, it is evident that the Egovehicle crashes into the NPC vehicle under certain conditions with perception_mode:=camera_lidar_fusion. I suspect that this occurs because the Egovehicle, upon detecting the car in front, cannot process the information in time to decelerate.

Question

1. Is this crash scenario caused by the simulator itself? I have observed that when the Egovehicle's speed is not high, the crash does not occur.

2. Does Autoware prioritize stopping for traffic lights or obstacle stops more? Are there other factors that cause the Egovehicle to stop or decelerate?

Any insights or suggestions on the reasons behind this scenario

Evidence

2024-07-26.13-38-14.mp4

2024-07-26.13-38-14.mp4

2024-07-26.13-38-14.mp4

In addition to the videos, here are screenshots showing the process CPU usage and the output from nvidia-smi:

[maxime-clem]

From the first video it looks like a detection issue. The obstacle points (Sensing) or the detection object (Perception) appear very late.
The stop itself is triggered correctly but it is a bit too gentle. Some parameters on the planning side could be tuned to allow a stronger brake.

[ExitedState]

Thank you. It did not crash when I increased the parameters at ~/autoware/src/launcher/autoware_launch/autoware_launch/config/planning/scenario_planning/common/common.param.yaml

Could you explain more about min_acc and max_acc parameters? Specifically, what is the difference between the parameters for normal driving constraints and the ones for constraints to be observed?
Additionally, why does Autoware not dynamically adjust these parameters (such as deceleration) for safety in emergency situations?

[maxime-clem]

constraints to be observed

These constraints are never violated by the planning modules.

constraints param for normal driving

Unless there is an emergency, the planning modules will try to respect these constraints.