Skip to content

Commit

Permalink
docs(lane_change): object filtering description (#9947)
Browse files Browse the repository at this point in the history 
* docs(lane_change): object filtering description

Signed-off-by: Zulfaqar Azmi <[email protected]>

* Move section up

Signed-off-by: Zulfaqar Azmi <[email protected]>

* Update planning/behavior_path_planner/autoware_behavior_path_lane_change_module/README.md

Co-authored-by: mkquda <[email protected]>

* Update planning/behavior_path_planner/autoware_behavior_path_lane_change_module/README.md

Co-authored-by: mkquda <[email protected]>

---------

Signed-off-by: Zulfaqar Azmi <[email protected]>
Co-authored-by: mkquda <[email protected]>
  • Loading branch information
@zulfaqar-azmi-t4 @mkquda
zulfaqar-azmi-t4 and mkquda authored Jan 21, 2025
1 parent 037c315 commit b7b3dd6
Showing 1 changed file with 128 additions and 158 deletions.
286 changes: 128 additions & 158 deletions planning/behavior_path_planner/autoware_behavior_path_lane_change_module/README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -279,6 +279,134 @@ Within this range, we sample the lateral acceleration for the ego vehicle. Simil
lateral_acceleration_resolution = (maximum_lateral_acceleration - minimum_lateral_acceleration) / lateral_acceleration_sampling_num
```

#### Object filtering

Before performing safety checks, predicted objects are categorized based on their current pose and behavior at the time. These categories help determine how each object impacts the lane change process and guide the safety evaluation.

The predicted objects are divided into four main categories:

- **Target Lane Leading**: Objects that overlap with the target lane and are in front of the ego vehicle. This category is further divided into three subcategories:
- Moving: Objects with a velocity above a certain threshold.
- Stopped: Stationary objects within the target lane.
- Stopped at Bound: Objects outside the target lane but close to its boundaries.
- **Target Lane Trailing**: Objects that overlap with the target lane or any lanes preceding the target lane. Only moving vehicles are included in this category.
- **Current Lane**: Objects in front of the ego vehicle in the ego vehicle's current lane.
- **Others**: Any objects not classified into the above categories.

![object lanes](./images/lane_objects.drawio.svg)

Furthermore, for **Target Lane Leading** and **Current Lane** objects, only those positioned within the lane boundary or before the goal position are considered. Objects exceeding the end of the lane or the goal position are classified as **Others**.

Once objects are filtered into their respective categories, they are sorted by distance closest to the ego vehicle to farthest.

The following diagram illustrates the filtering process,

```plantuml
@startuml
skinparam defaultTextAlignment center
skinparam backgroundColor #WHITE
title Filter Objects main flowchart
start
:Filter predicted objects by class;
note left: Remove objects whose classes are\nnot specified in the <color:red>**target_object**</color> parameter.
:Filter oncoming predicted objects;
note left: Compare ego's current pose and object's current pose,\nand filter out any object whose yaw difference exceeds\n<color:red>**collision_check.th_incoming_object_yaw**</color>;
:Transform predicted objects to extended predicted objects;
group "Filter target lane objects" {
if (Object's lateral distance\nfrom the centerline\nis more than\nhalf of ego's width?) then (TRUE)
if (Object's current pose\nis before the goal or\nthe end of the lane) then (TRUE)
if (Object overlaps target lane?) then (TRUE)
#LightGreen:To SUBPROCESS "Separate object based on its behavior in target lane";
if(can separate object in SUBPROCESS) then (TRUE)
stop
endif
else (FALSE)
if (Object is moving\nAND\nis a vehicle class\nAND\npath overlaps target lane?) then (TRUE)
#LightGreen:To SUBPROCESS "Separate object based on its behavior in target lane";
if(can separate object in SUBPROCESS) then (TRUE)
stop
endif
endif
endif
endif
#LightPink:From SUBPROCESS "Separate object based on its behavior in target lane";
if (Object is in expanded target lane\nAND\nis stopped\nAND\nin front of ego?) then (TRUE)
:Add object to <color:blue>**filtered_object.leading_objects.stopped_at_bound**</color> list;
stop
endif
else (FALSE)
if (Object overlaps preceding target lanes?) then (TRUE)
:Add object to <color:blue>**filtered_object.trailing_objects**</color> list;
stop
endif
endif
}
if (Object is in front of ego\nAND\nis before terminal\nAND\noverlaps current lane?) then (TRUE)
:Add object to <color:orange>**filtered_object.current_lane**</color> list;
stop
else (FALSE)
:Add object to <color:magenta>**filtered_object.others**</color> list;
endif
stop
group Target Lane Objects Subprocess #LightYellow
start
if (Object is behind ego?) then (TRUE)
if (Object is moving?) then (TRUE)
:Add object to <color:blue>**filtered_object.trailing_objects**</color> list;
#LightGreen:Can separate object;
stop
else (FALSE)
#LightPink:Cant separate proceed with other checks;
stop
endif
else (FALSE)
if (Object is moving?) then (TRUE)
:Add object to <color:blue>**filtered_object.leading_objects.moving**</color> list;
else (FALSE)
:Add object to <color:blue>**filtered_object.leading_objects.stopped**</color> list;
endif
#LightGreen:Can separate object;
stop
endif
@enduml
```

!!! note

As shown in the flowchart, oncoming objects are also filtered out. The filtering process considers the difference between the current orientation of the ego vehicle and that of the object. However, depending on the map's geometry, a certain threshold may need to be allowed. This threshold can be configured using the parameter collision_check.th_incoming_object_yaw.

!!! note

The **Target Lane Leading's Stopped at Boundary** objects are detected using the expanded area of the target lane beyond its original boundaries. The parameters `lane_expansion.left_offset` and `lane_expansion.right_offset` can be configured to adjust the expanded width.

<div align="center">
<table>
<tr>
<td>
<div style="text-align: center;">
<div style="color: black; font-size: 20px; margin-bottom: 10px;">Without Lane Expansion</div>
<img src="./images/lane_change-lane_expansion-without.png" alt="Without lane expansion">
</div>
</td>
<td>
<div style="text-align: center;">
<div style="color: black; font-size: 20px; margin-bottom: 10px;">With Lane Expansion</div>
<img src="./images/lane_change-lane_expansion-with.png" alt="With lane expansion">
</div>
</td>
</tr>
</table>
</div>

#### Candidate Path's validity check

A candidate path is considered valid if it meets the following criteria:
Expand Down Expand Up @@ -424,166 +552,8 @@ See [safety check utils explanation](../autoware_behavior_path_planner_common/do

First, we divide the target objects into obstacles in the target lane, obstacles in the current lane, and obstacles in other lanes. Target lane indicates the lane that the ego vehicle is going to reach after the lane change and current lane mean the current lane where the ego vehicle is following before the lane change. Other lanes are lanes that do not belong to the target and current lanes. The following picture describes objects on each lane. Note that users can remove objects either on current and other lanes from safety check by changing the flag, which are `check_objects_on_current_lanes` and `check_objects_on_other_lanes`.

![object lanes](./images/lane_objects.drawio.svg)

Furthermore, to change lanes behind a vehicle waiting at a traffic light, we skip the safety check for the stopping vehicles near the traffic light. The explanation for parked car detection is written in [documentation for avoidance module](../autoware_behavior_path_static_obstacle_avoidance_module/README.md).

The detection area for the target lane can be expanded beyond its original boundaries to enable detection of objects that are outside the target lane's limits.

<div align="center">
<table>
<tr>
<td>
<div style="text-align: center;">
<div style="color: black; font-size: 20px; margin-bottom: 10px;">Without Lane Expansion</div>
<img src="./images/lane_change-lane_expansion-without.png" alt="Without lane expansion">
</div>
</td>
<td>
<div style="text-align: center;">
<div style="color: black; font-size: 20px; margin-bottom: 10px;">With Lane Expansion</div>
<img src="./images/lane_change-lane_expansion-with.png" alt="With lane expansion">
</div>
</td>
</tr>
</table>
</div>

##### Object filtering

```plantuml
@startuml
skinparam defaultTextAlignment center
skinparam backgroundColor #WHITE
title NormalLaneChange::filterObjects Method Execution Flow
start
group "Filter Objects by Class" {
while (has not finished iterating through predicted object list) is (TRUE)
if (current object type != param.object_types_to_check?) then (TRUE)
#LightPink:Remove current object;
else (FALSE)
:Keep current object;
endif
end while
end group
if (predicted object list is empty?) then (TRUE)
:Return empty result;
stop
else (FALSE)
endif
group "Filter Oncoming Objects" #PowderBlue {
while (has not finished iterating through predicted object list?) is (TRUE)
if (object's yaw with reference to ego's yaw difference < 90 degree?) then (TRUE)
:Keep current object;
else (FALSE)
if (object is stopping?) then (TRUE)
:Keep current object;
else (FALSE)
#LightPink:Remove current object;
endif
endif
endwhile
end group
if (predicted object list is empty?) then (TRUE)
:Return empty result;
stop
else (FALSE)
endif
group "Filter Objects By Lanelets" #LightGreen {
while (has not finished iterating through predicted object list) is (TRUE)
:Calculate lateral distance diff;
if (Object in target lane polygon?) then (TRUE)
if (lateral distance diff > half of ego's width?) then (TRUE)
if (Object's physical position is before terminal point?) then (TRUE)
:Add to target_lane_objects;
else (FALSE)
endif
else (FALSE)
endif
else (FALSE)
endif
if (Object overlaps with backward target lanes?) then (TRUE)
:Add to target_lane_objects;
else (FALSE)
if (Object in current lane polygon?) then (TRUE)
:Add to current_lane_objects;
else (FALSE)
:Add to other_lane_objects;
endif
endif
end while
:Return target lanes object, current lanes object and other lanes object;
end group
:Generate path from current lanes;
if (path empty?) then (TRUE)
:Return empty result;
stop
else (FALSE)
endif
group "Filter Target Lanes' objects" #LightCyan {
while (has not finished iterating through target lanes' object list) is (TRUE)
if(velocity is within threshold?) then (TRUE)
:Keep current object;
else (FALSE)
if(object is ahead of ego?) then (TRUE)
:keep current object;
else (FALSE)
#LightPink:remove current object;
endif
endif
endwhile
end group
group "Filter Current Lanes' objects" #LightYellow {
while (has not finished iterating through current lanes' object list) is (TRUE)
if(velocity is within threshold?) then (TRUE)
if(object is ahead of ego?) then (TRUE)
:keep current object;
else (FALSE)
#LightPink:remove current object;
endif
else (FALSE)
#LightPink:remove current object;
endif
endwhile
end group
group "Filter Other Lanes' objects" #Lavender {
while (has not finished iterating through other lanes' object list) is (TRUE)
if(velocity is within threshold?) then (TRUE)
if(object is ahead of ego?) then (TRUE)
:keep current object;
else (FALSE)
#LightPink:remove current object;
endif
else (FALSE)
#LightPink:remove current object;
endif
endwhile
end group
:Transform the objects into extended predicted object and return them as lane_change_target_objects;
stop
@enduml
```

##### Collision check in prepare phase

The ego vehicle may need to secure ample inter-vehicle distance ahead of the target vehicle before attempting a lane change. The flag `enable_collision_check_at_prepare_phase` can be enabled to gain this behavior. The following image illustrates the differences between the `false` and `true` cases.
Expand Down

0 comments on commit b7b3dd6

Please sign in to comment.