Trajectory Validator

Purpose/Role

This package provides a pluginlib-based C++ library for evaluating candidate trajectories against a configurable set of safety and traffic-rule filters. Each filter plugin receives a world-state context snapshot (ValidatorContext) and returns a feasibility verdict for a single trajectory. Trajectories rejected by any enforced plugin are removed from the output set. The library is embedded in autoware_trajectory_selector.

Algorithm Overview

For each input CandidateTrajectories message the library runs every loaded plugin against every trajectory:

1. Plugin evaluation: each plugin's is_feasible() is called with the trajectory points and the current ValidatorContext (odometry, predicted objects, acceleration, HD map, traffic light states).
2. Feasibility decision: a trajectory survives if every plugin listed in filter_names returns is_feasible = true. Plugins listed in shadow_mode_filter_names are evaluated and reported but never remove a trajectory.
3. Diagnostics: the wrapper publishes an Autoware diagnostics status — OK if all trajectories pass all enforced plugins, WARN if at least one is rejected, ERROR if none survive.

Built-in Plugins

Class name	Category	Description
safety::VehicleConstraintFilter	safety	Rejects trajectories that exceed maximum speed, acceleration, deceleration, steering angle, or steering rate.
safety::UncrossableBoundaryDepartureFilter	safety	Rejects trajectories whose footprint crosses an uncrossable map boundary (e.g., road borders).
traffic_rule::TrafficLightFilter	traffic_rule	Rejects trajectories that cross red, amber, or configured unknown traffic-light stop lines.

Interface

This package is a C++ library. The topics below are subscribed to by the hosting node (trajectory_selector_node) and passed to the validator as a ValidatorContext.

Topics

Direction	Topic name	Message Type	Description
Subscriber	~/input/odometry	nav_msgs/msg/Odometry	Current ego pose and velocity (mandatory)
Subscriber	~/input/acceleration	geometry_msgs/msg/AccelWithCovarianceStamped	Current ego acceleration (mandatory)
Subscriber	~/input/objects	autoware_perception_msgs/msg/PredictedObjects	Surrounding dynamic obstacles (mandatory)
Subscriber	~/input/traffic_signals	autoware_perception_msgs/msg/TrafficLightGroupArray	Traffic light states (optional; absent data does not block validation)
Subscriber	~/input/route	autoware_planning_msgs/msg/LaneletRoute	Route with the lanelets followed by ego (only required by the TrafficLightFilter)
Subscriber	~/input/lanelet2_map	autoware_map_msgs/msg/LaneletMapBin	HD map loaded once at startup (transient local QoS; mandatory)
Publisher	~/debug/validation_reports	autoware_trajectory_validator/msg/ValidationReportArray	Per-trajectory validation verdict and per-metric values
Publisher	~/debug/markers/<plugin_name>	visualization_msgs/msg/MarkerArray	Per-plugin debug visualization markers
Publisher	~/debug/plugin_report_text	visualization_msgs/msg/MarkerArray	Text overlay summarizing how many paths each plugin filtered
Publisher	~/debug/processing_time_ms	autoware_internal_debug_msgs/msg/Float64Stamped	Total validator processing time [ms]
Publisher	~/debug/<plugin_name>/processing_time_ms	autoware_internal_debug_msgs/msg/Float64Stamped	Per-plugin processing time [ms]
Publisher	~/debug/processing_time_text	autoware_internal_debug_msgs/msg/StringStamped	Human-readable processing time breakdown

Parameters

Name	Type	Description	Default	Range
filter_names	array	List of validator plugin class names to enforce. Trajectories rejected by any plugin in this list are removed from the output.	['']	N/A
shadow_mode_filter_names	array	List of validator plugin class names that run in shadow mode: results are evaluated, logged, and reported via diagnostics but do not affect the output trajectory set.	['']	N/A
vehicle_constraint.max_speed	float	Maximum allowed longitudinal speed [m/s].	16.7	≥0.0
vehicle_constraint.max_acceleration	float	Maximum allowed longitudinal acceleration [m/s²].	5.0	≥0.0
vehicle_constraint.max_deceleration	float	Maximum allowed deceleration magnitude [m/s²] (positive value representing deceleration).	5.0	≥0.0
vehicle_constraint.max_steering_angle	float	Maximum allowed front-wheel steering angle [rad].	0.8	≥0.0
vehicle_constraint.max_steering_rate	float	Maximum allowed steering rate [rad/s].	0.3	≥0.0
traffic_light.deceleration_limit	float	Deceleration limit used to decide whether ego can safely stop before an amber stop line [m/s²]. A trajectory crossing an amber light is rejected if ego could stop within this deceleration.	2.8	N/A
traffic_light.jerk_limit	float	Jerk limit used together with deceleration_limit to compute the minimum stopping distance [m/s³].	5.0	N/A
traffic_light.delay_response_time	float	Driver/system response delay used to estimate the earliest possible start of braking [s].	0.5	≥0.0
traffic_light.crossing_time_limit	float	A trajectory crossing an amber light is rejected unless ego can fully clear the stop line within this time [s].	2.75	≥0.0
traffic_light.treat_amber_light_as_red_light	boolean	When true, amber lights are treated identically to red lights (no dilemma-zone logic is applied).	True	N/A
traffic_light.treat_unknown_light_as_red_light	boolean	When true, UNKNOWN lights are treated identically to red lights.	False	N/A
traffic_light.stop_overshoot_margin	float	Maximum allowed distance between the trajectory's stop point and the stop line for the trajectory to be considered as stopping correctly [m].	0.5	≥0.0
traffic_light.allow_if_cannot_stop_distance	float	allow crossing a stop line within this distance if ego cannot stop before the line plus its margin [m].	0.0	≥0.0
traffic_light.stable_duration_threshold_red	float	Minimum duration a RED light must be seen before it is considered active (only when ego is moving) [s].	0.0	≥0.0
traffic_light.stable_duration_threshold_amber	float	Minimum duration an AMBER light must be seen before it is considered active (only when ego is moving) [s].	0.0	≥0.0
traffic_light.stable_duration_threshold_unknown	float	Minimum duration an UNKNOWN light must be seen before it is considered active (only when ego is moving) [s].	0.0	≥0.0
traffic_light.amber_rejection_hysteresis_duration	float	Duration to persist an AMBER rejection state to prevent flipping [s].	0.0	≥0.0
traffic_light.ego_stopped_velocity_threshold	float	Velocity threshold below which stability and hysteresis filters are bypassed [m/s].	0.01	≥0.0
checked_trajectory_length.deceleration_limit	float	Deceleration limit used to compute the trajectory length scanned for traffic light stop lines [m/s²].	2.0	N/A
checked_trajectory_length.jerk_limit	float	Jerk limit used together with deceleration_limit to compute the trajectory scan length [m/s³].	2.0	N/A
boundary_departure.lateral_margin_m	float	Lateral clearance margin from an uncrossable boundary [m]. Trajectories whose footprint comes within this margin are rejected.	0.01	N/A
boundary_departure.longitudinal_margin_m	float	Longitudinal look-ahead margin applied when projecting the vehicle footprint along the trajectory [m].	1.0	N/A
boundary_departure.max_deceleration_mps2	float	Maximum deceleration (negative value) used to compute whether ego can stop before the boundary [m/s²].	-4.0	N/A
boundary_departure.max_jerk_mps3	float	Maximum jerk (negative value) used together with max_deceleration_mps2 for the stopping distance estimate [m/s³].	-5.0	N/A
boundary_departure.brake_delay_s	float	Braking response delay [s].	1.0	≥0.0
boundary_departure.time_to_departure_cutoff_s	float	Time-to-boundary-crossing threshold below which a trajectory is flagged as departing [s].	2.0	≥0.0
boundary_departure.on_time_buffer_s	float	Hysteresis buffer before activating boundary-departure detection [s].	0.15	≥0.0
boundary_departure.off_time_buffer_s	float	Hysteresis buffer before deactivating boundary-departure detection [s].	0.15	≥0.0
boundary_departure.enable_developer_marker	boolean	Publish diagnostic visualization markers intended for developer use.	True	N/A
boundary_departure.boundary_types	array	List of lanelet boundary type strings to treat as uncrossable (e.g., 'road_border').	['road_border']	N/A