Surround Obstacle Checker

Purpose

This module subscribes required data (ego-pose, obstacles, etc), and publishes zero velocity limit to keep stopping if any of stop conditions are satisfied.

Inner-workings / Algorithms

Flow chart

@startuml

title surround obstacle checker
start

if (Check state) then
  :State::STOP;

  if (Is stop required?) then (yes)
  else (no)
    :Clear velocity limit;
    :Set state to State::PASS;
  endif

else
  :State::PASS;

  if (Is stop required?) then (yes)
    :Set velocity limit;
    :Set state to State::STOP;
  else (no)
  endif

endif

:Publish stop reason;

stop
@enduml

Algorithms

Check data

Check that surround_obstacle_checker receives no ground pointcloud, dynamic objects and current velocity data.

Get distance to nearest object

Calculate distance between ego vehicle and the nearest object. In this function, it calculates the minimum distance between the polygon of ego vehicle and all points in pointclouds and the polygons of dynamic objects.

Stop requirement

If it satisfies all following conditions, it plans stopping.

• Ego vehicle is stopped
• It satisfies any following conditions
  1. The distance to nearest obstacle satisfies following conditions
     • If state is State::PASS, the distance is less than surround_check_distance
     • If state is State::STOP, the distance is less than surround_check_recover_distance
  2. If it does not satisfies the condition in 1, elapsed time from the time it satisfies the condition in 1 is less than state_clear_time

States

To prevent chattering, surround_obstacle_checker manages two states. As mentioned in stop condition section, it prevents chattering by changing threshold to find surround obstacle depending on the states.

• State::PASS : Stop planning is released
• State::STOP ：While stop planning

Inputs / Outputs

Input

Output

Parameters

{{ json_to_markdown(“planning/surround_obstacle_checker/schema/surround_obstacle_checker.schema.json”) }}

Assumptions / Known limits

To perform stop planning, it is necessary to get obstacle pointclouds data. Hence, it does not plan stopping if the obstacle is in blind spot.