Run Out

Role

run_out is the module that decelerates and stops for dynamic obstacles such as pedestrians and bicycles.

Activation Timing

This module is activated if launch_run_out becomes true

Inner-workings / Algorithms

Flow chart

@startuml
title modifyPathVelocity
start

partition "Preprocess path" {
:Calculate the expected target velocity for ego vehicle;
:Extend path;
:Trim path from ego position;
}

partition "Preprocess obstacles" {
:Create data of abstracted dynamic obstacles;
:Exclude obstacles outside of partition lanelet;
}

partition "Collision_detection" {
:Detect collision with dynamic obstacles;
}

partition "Insert velocity" {
:Insert velocity to decelerate for obstacles;

:Limit velocity with specified jerk and acc limit;
}
stop
@enduml

Preprocess path

Calculate the expected target velocity for ego vehicle

Calculate the expected target velocity for the ego vehicle path to calculate time to collision with obstacles more precisely. The expected target velocity is calculated with motion velocity smoother module by using current velocity, current acceleration and velocity limits directed by the map and external API.

Extend the path

The path is extended by the length of base link to front to consider obstacles after the goal.

Trim path from ego position

The path is trimmed from ego position to a certain distance to reduce calculation time. Trimmed distance is specified by parameter of detection_distance.

Preprocess obstacles

Create data of abstracted dynamic obstacle

This module can handle multiple types of obstacles by creating abstracted dynamic obstacle data layer. Currently we have 3 types of detection method (Object, ObjectWithoutPath, Points) to create abstracted obstacle data.

Abstracted dynamic obstacle

Abstracted obstacle data has following information.

Enter the maximum/minimum velocity of the object as a parameter, adding enough margin to the expected velocity. This parameter is used to create polygons for collision detection.

Future work: Determine the maximum/minimum velocity from the estimated velocity with covariance of the object

3 types of detection method

We have 3 types of detection method to meet different safety and availability requirements. The characteristics of them are shown in the table below. Method of Object has high availability (less false positive) because it detects only objects whose predicted path is on the lane. However, sometimes it is not safe because perception may fail to detect obstacles or generate incorrect predicted paths. On the other hand, method of Points has high safety (less false negative) because it uses pointcloud as input. Since points don’t have a predicted path, the path that moves in the direction normal to the path of ego vehicle is considered to be the predicted path of abstracted dynamic obstacle data. However, without proper adjustment of filter of points, it may detect a lot of points and it will result in very low availability. Method of ObjectWithoutPath has the characteristics of an intermediate of Object and Points.

File truncated at 100 lines see the full file