Skip to content

synchronized OGM fusion#

For simplicity, we use OGM as the meaning of the occupancy grid map.

This package is used to fuse the OGMs from synchronized sensors. Especially for the lidar.

Here shows the example OGM for the this synchronized OGM fusion.

left lidar OGM right lidar OGM top lidar OGM
left right top

OGM fusion with asynchronous sensor outputs is not suitable for this package. Asynchronous OGM fusion is under construction.

Processing flow#

The processing flow of this package is shown in the following figure.

data_flow

  • Single Frame Fusion
    • Single frame fusion means that the OGMs from synchronized sensors are fused in a certain time frame \(t=t_n\).
  • Multi Frame Fusion
    • In the multi frame fusion process, current fused single frame OGM in \(t_n\) is fused with the previous fused single frame OGM in \(t_{n-1}\).

I/O#

Input topic name Type Description
input_ogm_topics list of nav_msgs::msg::OccupancyGrid List of input topics for Occupancy Grid Maps. This parameter is given in list, so
Output topic name Type Description
~/output/occupancy_grid_map nav_msgs::msg::OccupancyGrid Output topic name of the fused Occupancy Grid Map.
~/debug/single_frame_map nav_msgs::msg::OccupancyGrid (debug topic) Output topic name of the single frame fused Occupancy Grid Map.

Parameters#

Synchronized OGM fusion node parameters are shown in the following table. Main parameters to be considered in the fusion node is shown as bold.

Ros param name Sample value Description
input_ogm_topics ["topic1", "topic2"] List of input topics for Occupancy Grid Maps
input_ogm_reliabilities [0.8, 0.2] Weights for the reliability of each input topic
fusion_method "overwrite" Method of fusion ("overwrite", "log-odds", "dempster-shafer")
match_threshold_sec 0.01 Matching threshold in milliseconds
timeout_sec 0.1 Timeout duration in seconds
input_offset_sec [0.0, 0.0] Offset time in seconds for each input topic
mapframe "map" Frame name for the fused map
baselink_frame "base_link" Frame name for the base link
gridmap_origin_frame "base_link" Frame name for the origin of the grid map
fusion_map_length_x 100.0 Length of the fused map along the X-axis
fusion_map_length_y 100.0 Length of the fused map along the Y-axis
fusion_map_resolution 0.5 Resolution of the fused map

Since this node assumes that the OGMs from synchronized sensors are generated in the same time, we need to tune the match_threshold_sec, timeout_sec and input_offset_sec parameters to successfully fuse the OGMs.

Fusion methods#

For the single frame fusion, the following fusion methods are supported.

Fusion Method in parameter Description
overwrite The value of the cell in the fused OGM is overwritten by the value of the cell in the OGM with the highest priority.
We set priority as Occupied > Free > Unknown.
log-odds The value of the cell in the fused OGM is calculated by the log-odds ratio method, which is known as a Bayesian fusion method.
The log-odds of a probability \(p\) can be written as \(l_p = \log(\frac{p}{1-p})\).
And the fused log-odds is calculated by the sum of log-odds. \(l_f = \Sigma l_p\)
dempster-shafer The value of the cell in the fused OGM is calculated by the Dempster-Shafer theory[1]. This is also popular method to handle multiple evidences. This package applied conflict escape logic in [2] for the performance. See references for the algorithm details.

For the multi frame fusion, currently only supporting log-odds fusion method.

How to use#

launch fusion node#

The minimum node launch will be like the following.

<?xml version="1.0"?>
<launch>
<arg name="output_topic" default="~/output/occupancy_grid_map"/>
<arg name="fusion_node_param_path" default="$(find-pkg-share autoware_probabilistic_occupancy_grid_map)/config/synchronized_grid_map_fusion_node.param.yaml"/>

<node name="synchronized_grid_map_fusion_node" exec="synchronized_grid_map_fusion_node" pkg="autoware_probabilistic_occupancy_grid_map" output="screen">
  <remap from="~/output/occupancy_grid_map" to="$(var output_topic)"/>
  <param from="$(var fusion_node_param_path)"/>
</node>
</launch>

(Optional) Generate OGMs in each sensor frame#

You need to generate OGMs in each sensor frame before achieving grid map fusion.

autoware_probabilistic_occupancy_grid_map package supports to generate OGMs for the each from the point cloud data.

Example launch.xml (click to expand) 
<include file="$(find-pkg-share tier4_perception_launch)/launch/occupancy_grid_map/probabilistic_occupancy_grid_map.launch.xml">
    <arg name="input/obstacle_pointcloud" value="/perception/obstacle_segmentation/single_frame/pointcloud"/>
    <arg name="input/raw_pointcloud" value="/sensing/lidar/right/outlier_filtered/pointcloud_synchronized"/>
    <arg name="output" value="/perception/occupancy_grid_map/right_lidar/map"/>
    <arg name="map_frame" value="base_link"/>
    <arg name="scan_origin" value="velodyne_right"/>
    <arg name="use_intra_process" value="true"/>
    <arg name="use_multithread" value="true"/>
    <arg name="use_pointcloud_container" value="$(var use_pointcloud_container)"/>
    <arg name="pointcloud_container_name" value="$(var pointcloud_container_name)"/>
    <arg name="method" value="pointcloud_based_occupancy_grid_map"/>
    <arg name="param_file" value="$(find-pkg-share autoware_probabilistic_occupancy_grid_map)/config/pointcloud_based_occupancy_grid_map_fusion.param.yaml"/>
</include>


The minimum parameter for the OGM generation in each frame is shown in the following table.

|Parameter|Description|
|--|--|
|`input/obstacle_pointcloud`| The input point cloud data for the OGM generation. This point cloud data should be the point cloud data which is segmented as the obstacle.|
|`input/raw_pointcloud`| The input point cloud data for the OGM generation. This point cloud data should be the point cloud data which is not segmented as the obstacle. |
|`output`| The output topic of the OGM. |
|`map_frame`| The tf frame for the OGM center origin. |
|`scan_origin`| The tf frame for the sensor origin. |
|`method`| The method for the OGM generation. Currently we support `pointcloud_based_occupancy_grid_map` and `laser_scan_based_occupancy_grid_map`. The pointcloud based method is recommended. |
|`param_file`| The parameter file for the OGM generation. See [example parameter file](config/pointcloud_based_occupancy_grid_map_for_fusion.param.yaml) |


We recommend to use same map_frame, size and resolutions for the OGMs from synchronized sensors.
Also, remember to set enable_single_frame_mode and filter_obstacle_pointcloud_by_raw_pointcloud to true in the autoware_probabilistic_occupancy_grid_map package (you do not need to set these parameters if you use the above example config file).


Run both OGM generation node and fusion node#

We prepared the launch file to run both OGM generation node and fusion node in grid_map_fusion_with_synchronized_pointclouds.launch.py

You can include this launch file like the following.

<include file="$(find-pkg-share autoware_probabilistic_occupancy_grid_map)/launch/grid_map_fusion_with_synchronized_pointclouds.launch.py">
  <arg name="output" value="/perception/occupancy_grid_map/fusion/map"/>
  <arg name="use_intra_process" value="true"/>
  <arg name="use_multithread" value="true"/>
  <arg name="use_pointcloud_container" value="$(var use_pointcloud_container)"/>
  <arg name="pointcloud_container_name" value="$(var pointcloud_container_name)"/>
  <arg name="method" value="pointcloud_based_occupancy_grid_map"/>
  <arg name="fusion_config_file" value="$(var fusion_config_file)"/>
  <arg name="ogm_config_file" value="$(var ogm_config_file)"/>
</include>

The minimum parameter for the launch file is shown in the following table.

Parameter Description
output The output topic of the finally fused OGM.
method The method for the OGM generation. Currently we support pointcloud_based_occupancy_grid_map and laser_scan_based_occupancy_grid_map. The pointcloud based method is recommended.
fusion_config_file The parameter file for the grid map fusion. See example parameter file
ogm_config_file The parameter file for the OGM generation. See example parameter file

References#