Rear Collision Checker#

The rear_collision_checker is a plugin module of the autoware_planning_validator node. It validates the planned trajectory by verifying that it does not lead to a collision with other road users (primarily vehicles, but also pedestrian/cyclists/motorcycles) approaching from lateral/rear directions. In particular, it is designed to detect potential rear-end collisions and entrapment risks when the ego vehicle is making a right/left turn or merging into an adjacent lane.

Inner Workings#

The module operates by:

Identifying conflict lanes
- Uses the current route and lanelet topology to determine lanes that the ego vehicle’s trajectory may intersect with during right/left turns or lane changes.
- Focuses on lanes where vehicles, bicycles, or motorcycles could approach from behind or from a lateral direction, posing a potential rear-end or entrapment risk.
- Considers the turn_direction of the approaching lanelets to filter out irrelevant lanes.
Filtering perception data
- Subscribes to obstacle pointcloud and first filters by the conflict region, keeping only points that fall within relevant spatial bounds.
- Performs clustering on the filtered points to form obstacle candidates.
- Associates clusters with nearby lanelets and computes the nearest face of each obstacle along the lane direction.
- Applies configurable range gates (forward/backward, lateral, height) and basic outlier rejection to discard distant or irrelevant clusters.
Estimating motion
- For each selected object, determines its motion relative to the lane direction.
- Estimates speed and direction based on frame-to-frame position changes along the lane’s centerline.
- Maintains a per-lane tracking list to stabilize velocity estimation and reduce noise from perception flicker.
Calculating metrics
- Depending on the selected metric (e.g., TTC, RSS), the calculation method and decision logic differ. The chosen metric is computed, and the result is used to determine whether there is a potential collision risk.
Judging collision risk
- Evaluates the selected metric results against predefined thresholds to determine whether a collision is likely.
- Applies hysteresis logic:
  - on_time_buffer: Hazardous state must persist for this duration before marking the trajectory unsafe.
  - off_time_buffer: Safe state must persist for this duration before clearing the unsafe flag.
- If check_on_unstoppable=false, the module will skip collision warnings in cases where the ego vehicle cannot realistically stop before the conflict area (to prevent unnecessary alerts when stopping is already infeasible).

Flowchart#

WIP

General Parameters#

Name	Unit	Type	Description	Default value
`on_time_buffer`	[s]	double	Time buffer before enabling detection after a relevant condition is met	0.5
`off_time_buffer`	[s]	double	Time buffer before disabling detection after the condition clears	1.5
`check_on_unstoppable`	[-]	bool	If `true`, the module continues collision checking even when the ego vehicle cannot stop before entering a potential collision area, and outputs an ERROR if a collision risk is detected. If `false`, checking is skipped in such cases to avoid unnecessary warnings.	false

Pointcloud Preprocess#

Name	Unit	Type	Description	Default value
`pointcloud.range.dead_zone`	[m]	double	Distance in front of the ego vehicle ignored for collision detection	0.3
`pointcloud.range.buffer`	[m]	double	Additional margin around detection range	1.0
`pointcloud.crop_box_filter.x.max`	[m]	double	Maximum X coordinate for cropped detection range	30.0
`pointcloud.crop_box_filter.x.min`	[m]	double	Minimum X coordinate for cropped detection range	-100.0
`pointcloud.crop_box_filter.z.max`	[m]	double	Maximum Z coordinate for cropped detection range	-1.0
`pointcloud.crop_box_filter.z.min`	[m]	double	Minimum Z coordinate for cropped detection range	0.3
`pointcloud.voxel_grid_filter.x`	[m]	double	Voxel grid filter size in X direction	0.1
`pointcloud.voxel_grid_filter.y`	[m]	double	Voxel grid filter size in Y direction	0.1
`pointcloud.voxel_grid_filter.z`	[m]	double	Voxel grid filter size in Z direction	0.5
`pointcloud.clustering.cluster_tolerance`	[m]	double	Maximum distance between points to be considered part of the same cluster	0.15
`pointcloud.clustering.min_cluster_height`	[m]	double	Minimum height of a cluster to be considered valid	0.1
`pointcloud.clustering.min_cluster_size`	[points]	int	Minimum number of points in a valid cluster	5
`pointcloud.clustering.max_cluster_size`	[points]	int	Maximum number of points in a valid cluster	10000
`pointcloud.velocity_estimation.observation_time`	[s]	double	Time window used for velocity estimation	0.3
`pointcloud.velocity_estimation.max_acceleration`	[m/s^2]	double	Maximum allowed acceleration in velocity estimation	10.0
`pointcloud.latency`	[s]	double	Assumed system latency for point cloud processing	0.3

Object Filtering#

Name	Unit	Type	Description	Default value
`filter.min_velocity`	[m/s]	double	Minimum velocity threshold for objects to be considered moving	1.0
`filter.moving_time`	[s]	double	Minimum time duration an object must be moving to be considered as such	0.5

TTC#

Name	Unit	Type	Description	Default value
`time_to_collision.margin`	[s]	double	Additional margin added to time-to-collision calculation	2.0

Ego Behavior#

Name	Unit	Type	Description	Default value
`ego.reaction_time`	[s]	double	Reaction time of the ego vehicle	1.2
`ego.min_velocity`	[m/s]	double	Minimum considered velocity of the ego vehicle	1.38
`ego.max_velocity`	[m/s]	double	Maximum considered velocity of the ego vehicle	16.6
`ego.max_acceleration`	[m/s^2]	double	Maximum considered acceleration of the ego vehicle	1.5
`ego.max_deceleration`	[m/s^2]	double	Maximum considered deceleration of the ego vehicle (negative value)	-4.0
`ego.max_positive_jerk`	[m/s^3]	double	Maximum allowed positive jerk for the ego vehicle	5.0
`ego.max_negative_jerk`	[m/s^3]	double	Maximum allowed negative jerk for the ego vehicle	-5.0
`ego.nominal_deceleration`	[m/s^2]	double	Nominal deceleration used for calculations	-1.5
`ego.nominal_positive_jerk`	[m/s^3]	double	Nominal positive jerk used for calculations	0.6
`ego.nominal_negative_jerk`	[m/s^3]	double	Nominal negative jerk used for calculations	-0.6

Name	Unit	Type	Description	Default value
`blind_spot.lookahead_time`	[s]	double	Lookahead time for blind spot detection	4.0
`blind_spot.metric`	[-]	string	Metric used for blind spot detection (`ttc`, etc.)	ttc
`blind_spot.check.front`	[-]	bool	Whether to check for blind spot in the front	false
`blind_spot.check.left`	[-]	bool	Whether to check for blind spot on the left	true
`blind_spot.check.right`	[-]	bool	Whether to check for blind spot on the right	false
`blind_spot.check.yaw_th`	[rad]	double	Yaw threshold for blind spot detection	0.78
`blind_spot.offset.inner`	[m]	double	Inner offset for blind spot detection zone	0.1
`blind_spot.offset.outer`	[m]	double	Outer offset for blind spot detection zone	0.3
`blind_spot.participants.reaction_time`	[s]	double	Reaction time of participants in blind spot detection	1.2
`blind_spot.participants.max_velocity`	[m/s]	double	Maximum velocity of participants in blind spot detection	5.5
`blind_spot.participants.max_deceleration`	[m/s^2]	double	Maximum deceleration of participants in blind spot detection	-2.0

Collision Check For Adjacent Lane#

Name	Unit	Type	Description	Default value
`adjacent_lane.lookahead_time`	[s]	double	Lookahead time for adjacent lane collision detection	4.0
`adjacent_lane.metric`	[-]	string	Metric used for adjacent lane collision detection (`rss`, etc.)	rss
`adjacent_lane.check.front`	[-]	bool	Whether to check for adjacent lane collision in the front	true
`adjacent_lane.offset.left`	[m]	double	Left offset for adjacent lane collision detection	-0.5
`adjacent_lane.offset.right`	[m]	double	Right offset for adjacent lane collision detection	-0.5
`adjacent_lane.participants.reaction_time`	[s]	double	Reaction time of participants in adjacent lane collision detection	1.2
`adjacent_lane.participants.max_velocity`	[m/s]	double	Maximum velocity of participants in adjacent lane collision detection	16.6
`adjacent_lane.participants.max_deceleration`	[m/s^2]	double	Maximum deceleration of participants in adjacent lane collision detection	-2.0