Collision Detector

Purpose

This module publishes an ERROR diagnostic if a collision is detected with the current ego footprint.

Inner-workings / Algorithms

Flow chart

1. Check input data.
2. Filter dynamic objects.
3. Find nearest object and its distance to ego.
4. Publish an ERROR diagnostic depending on the recent collision detection results.

Algorithms

Check data

Check that collision_detector receives no ground pointcloud, dynamic objects.

Object Filtering

Recognition Assumptions

1. If the classification changes but it's considered the same object, the uuid does not change.
2. It's possible for the same uuid to be recognized after being lost for a few frames.
3. Once an object is determined to be excluded, it continues to be excluded for a certain period of time.

Filtering Process

1. Initial Recognition and Exclusion:

   • The system checks if a newly recognized object's classification is listed in nearby_object_type_filters.
   • If so, and the object is within the nearby_filter_radius, it is marked for exclusion.

2. New Object Determination:

   • An object is considered "new" based on its UUID.
   • If the UUID is not found in recent frame data, the object is treated as new.

3. Exclusion Mechanism:

   • Newly excluded objects are recorded by their UUID.
   • These objects continue to be excluded for a set period (keep_ignoring_time) as long as they maintain the classification specified in nearby_object_type_filters and remain within the nearby_filter_radius.

Get distance to nearest object

Calculate distance between ego vehicle and the nearest object. In this function, it calculates the minimum distance between the polygon of ego vehicle and all points in pointclouds and the polygons of dynamic objects. If the minimum distance is lower than the collision_distance parameter, then a collision is detected.

Time buffer and distance hysteresis

Before publishing an ERROR diagnostic, a collision must be detected for at least a duration set by the parameter time_buffer.on. Once an ERROR diagnostic is published, the time_buffer.off_distance_hysteresis parameter is used to make the ego footprint larger, making it easier to detect a collision. To stop publishing the ERROR diagnostic, no collision must be detected for at least a duration set by the parameter time_buffer.off.

Inputs / Outputs

Input

Name	Type	Description
/perception/obstacle_segmentation/pointcloud	sensor_msgs::msg::PointCloud2	Pointcloud of obstacles which the ego-vehicle should stop or avoid
/perception/object_recognition/objects	autoware_perception_msgs::msg::PredictedObjects	Dynamic objects
/tf	tf2_msgs::msg::TFMessage	TF
/tf_static	tf2_msgs::msg::TFMessage	TF static

Output

Name	Type	Description
/diagnostics	diagnostic_msgs::msg::DiagnosticArray	Diagnostics
~/debug_markers	visualization_msgs::msg::MarkerArray	Debug markers

Parameters

Name	Type	Description	Default value
use_pointcloud	bool	Use pointcloud as obstacle check	true
use_dynamic_object	bool	Use dynamic object as obstacle check	true
collision_distance	double	Distance threshold at which an object is considered a collision. [m]	0.15
nearby_filter_radius	double	Distance range for filtering objects. Objects within this radius are considered. [m]	5.0
keep_ignoring_time	double	Time to keep filtering objects that first appeared in the vicinity [sec]	10.0
nearby_object_type_filters	object of bool values	Specifies which object types to filter. Only objects with true value will be filtered.	{unknown: true, others: false}
ignore_behind_rear_axle	bool	If true, collisions detected behind the rear axle of the ego vehicle are ignored	true
time_buffer.on	double	[s] minimum consecutive detection time before triggering the ERROR diagnostic	0.2
time_buffer.off	double	[s] minimum consecutive time without collision detection (including the hysteresis) before releasing the ERROR diagnostic	5.0
time_buffer.off_distance_hysteresis	double	[m] extra distance used to detect collisions once the diagnostic is triggered	1.0

Assumptions / Known limits

• This module is based on surround_obstacle_checker