Intersection

Role

The intersection module is responsible for safely passing urban intersections by:

1. checking collisions with upcoming vehicles
2. recognizing the occluded area in the intersection
3. reacting to each color/shape of associated traffic lights

This module is designed to be agnostic to left-hand/right-hand traffic rules and work for crossroads, T-shape junctions, etc. Roundabout is not formally supported in this module.

Activation condition

This module is activated when the path contains the lanes with turn_direction tag. More precisely, if the lane_ids of the path contain the ids of those lanes, corresponding instances of intersection module are activated on each lane respectively.

Requirements/Limitations

• The HDMap needs to have the information of turn_direction tag (which should be one of straight, left, right) for all the lanes in intersections and right_of_way tag for specific lanes (refer to RightOfWay section for more details). See autoware_lanelet2_extension document for more detail.
• WIP(perception requirements/limitations)
• WIP(sensor visibility requirements/limitations)

Attention area

The attention area in the intersection is defined as the set of lanes that are conflicting with ego path and their preceding lanes up to common.attention_area_length meters. By default RightOfWay tag is not set, so the attention area covers all the conflicting lanes and its preceding lanes as shown in the first row. RightOfWay tag is used to rule out the lanes that each lane has priority given the traffic light relation and turn_direction priority. In the second row, purple lanes are set as the yield_lane of the ego_lane in the RightOfWay tag.

intersection_area, which is supposed to be defined on the HDMap, is an area converting the entire intersection.

In-phase/Anti-phase signal group

The terms “in-phase signal group” and “anti-phase signal group” are introduced to distinguish the lanes by the timing of traffic light regulation as shown in below figure.

The set of intersection lanes whose color is in sync with lane L1 is called the in-phase signal group of L1, and the set of remaining lanes is called the anti-phase signal group.

How-to/Why set RightOfWay tag

Ideally RightOfWay tag is unnecessary if ego has perfect knowledge of all traffic signal information because:

• it can distinguish which conflicting lanes should be checked because they are GREEN currently and possible collision occur with the vehicles on those lanes
• it can distinguish which conflicting lanes can be ignored because they are RED currently and there is no chance of collision with the vehicles on those lanes unless they violate the traffic rule

That allows ego to generate the attention area dynamically using the real time traffic signal information. However this ideal condition rarely holds unless the traffic signal information is provided through the infrastructure. Also there maybe be very complicated/bad intersection maps where multiple lanes overlap in a complex manner.

• If there is an perfect access to entire traffic light signal, then you can set common.use_map_right_of_way to false and there is no need to set RightOfWay tag on the map. The intersection module will generate the attention area by checking traffic signal and corresponding conflicting lanes. This feature is not implemented yet.
• If traffic signal information is not perfect, then set common.use_map_right_of_way to true. If you do not want to detect vehicles on the anti-phase signal group lanes, set them as yield_lane for ego lane.
• Even if there are no traffic lights if the intersection lanes are overlapped in a ugly manner, you may need to set RightOfWay tag. For example if adjacent intersection lanes of the same in-phase group are not sharing the boundary line and overlapped a little bit, you may need to set RightOfWay to each other for them in order to avoid unnecessary stop for vehicle on such unrelated lane.

To help the intersection module care only a set of limited lanes, RightOfWay tag needs to be properly set.

Following table shows an example of how to set yield_lanes to each lane in a intersection w/o traffic lights. Since it is not apparent how to uniquely determine signal phase group for a set of intersection lanes in geometric/topological manner, yield_lane needs to be set manually. Straight lanes with traffic lights are exceptionally handled to detect no lanes because commonly it has priority over all the other lanes, so no RightOfWay setting is required.

This setting gives the following attention_area configurations.

For complex/bad intersection map like the one illustrated below, additional RightOfWay setting maybe necessary.

The bad points are:

1. ego lane is overlapped with adjacent lane of the in-phase group. In this case you need to set this lane as yield_lane additionally because otherwise attention area is generated for its preceding lanes as well, which may cause unwanted stop.
2. ego lane is overlapped with unrelated lane. In this case the lane is right-turn only and there is no chance of collision in theory. But you need to set this lane as yield_lane additionally for the same reason as (1).

Possible stop lines

Following figure illustrates important positions used in the intersection module. Note that each solid line represents ego front line position and the corresponding dot represents the actual inserted stop point position for the vehicle frame, namely the center of the rear wheel.

To precisely calculate stop positions, the path is interpolated at the certain interval of common.path_interpolation_ds.

• closest_idx denotes the path point index which is closest to ego position.
• first_attention_stopline denotes the first path point where ego footprint intersects with the attention_area.
• If a stopline is associated with the intersection lane on the map, that line is used as default_stopline for collision detection. Otherwise the point which is common.default_stopline_margin meters behind first_attention_stopline is defined as default_stopline instead.
• occlusion_peeking_stopline is a bit ahead of first_attention_stopline as described later.
• occlusion_wo_tl_pass_judge_line is the first position where ego footprint intersects with the centerline of the first attention_area lane.

Target objects

For stuck vehicle detection and collision detection, this module checks car, bus, truck, trailer, motor cycle, and bicycle type objects.

Objects that satisfy all of the following conditions are considered as target objects (possible collision objects):

File truncated at 100 lines see the full file

Changelog for package autoware_behavior_velocity_intersection_module

0.47.0 (2025-08-11)

• style(pre-commit): update to clang-format-20 (#11088) Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

• fix(intersection): remove reference "traffic_light_id" as it is deprecated (#11008)

• fix(intersection): clamp collision stopline ip index (#11009)

• feat(intersection_occlusion): request approval when occluded without traffic light (#10947)

• feat(intersection_module): add infomation to PlanningFactor topic (#10835)

  • feat(intersection_module): add infomation to safety factor
  • feat(intersection_module): add planning_factor_interface_for_occlusion_
  • feat(intersection_module): add planning factor test case
  • some code fix
  • add planning factor test case for occlusion
  • fix comment typo
  • marked planning_factor_interface_for_occlusion_as maybe_unused.
  • fix safety_factors and detail
  • fix test case and add comment

  * fix planning factor detail when decision is safe ---------

• feat(intersection): adjust collision stopline position to be ahead of ego by braking distance (#10900)

• fix(intersection): stop at occlusion stopline and stop twice at intersection without traffic light (#10911) feat(intersection): stop at occlusion stopline and stop twice at intersection without traffic light

• feat(intersection_occlusion): adjustable occlusion wall position for intersection without traffic_light (#10868)

• Contributors: Kotakku, Mamoru Sobue, Mete Fatih Cırıt, Yuxuan Liu

0.46.0 (2025-06-20)

• Merge remote-tracking branch 'upstream/main' into tmp/TaikiYamada/bump_version_base
• fix(intersection): fix path lanelets generation (#10843)
• feat(intersection): consider brake feasiblity for stuck vehicle (#10742)
• feat(intersection): experimental overshoot in merging (#10544)
• Contributors: Mamoru Sobue, TaikiYamada4

0.45.0 (2025-05-22)

• Merge remote-tracking branch 'origin/main' into tmp/notbot/bump_version_base

• fix(intersection): interpolate object polygons precisely (#10626) fix(intersection): debug collision check

• fix(behavior_velocity_planner): add missing header (#10561) Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

• feat(behavior_velocity_planner): extend stop line to path bound (#10435)

  • extend stop line to path bound
  • style(pre-commit): autofix
  • change signature of stop line extension function
  • add tests for detection area module
  • fix test for no stopping area module
  • fix tests for traffic light module
  • include necessary header

  * fix test for detection area module ---------Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

• feat(behavior_velocity_planner): only wait for the required subscriptions (#10546)

• Contributors: Mamoru Sobue, Masaki Baba, Mitsuhiro Sakamoto, TaikiYamada4, Takayuki Murooka

File truncated at 100 lines see the full file