About | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange

NOTE: This package is unreleased at this point, the documentation is for Willow Garage internal use only.

Overview

This package contains the multi_table_node, which is a node that processes point clouds of type PointCloud2 to detect tables and objects, and generates good pick and place poses around the tables. Good pick poses are defined as those that are near high object density areas, and good place poses are those that are located far from table corners and near areas of low object density. The multi_table_node is invoked with a service call, typically from a high level application such as one implemented in SMACH.

ROS API

multi_table_node contains many parameters for tweaking many aspects of the table and object detection, as well as the manipulation pose generation. In addition, the node needs to connect to the navigation stack in order to test the reachability of possible manipulation poses.

ROS Parameters

Parameters

General

~processing_frame (string, default: "")

• The frame that incoming points will be converted to before being processed. Empty string means no conversion will take place.

~up_direction (double, default: 1.0)

• The direction of the positive z axis in the processing frame.

Table Detection

~table_z_filter_min, ~table_z_filter_max (double, default: 0.4, 1.25)

• The minimum and maximum heights for points to pass through the initial filter.

~table_detection_voxel_size (double, default: 0.01)

• The size of the bins when downsampling points before clustering.

~min_table_cluster_size (int, default: 300)

• The minimum number of points in a cluster for it to be considered a possible table.

~table_cluster_distance (double, default: 0.06)

• The threshold for clustering when trying to find tables in the point cloud.

~table_inlier_threshold (int, default: 300)

• The minimum number of inliers in the plane fit to be considered a likely table.

Object Detection

~object_z_filter_min, ~object_z_filter_max (double, default: 0.01, 0.50)

• The minimum and maximum heights above the table top (expressed in the table top frame) that will be considered for object detection.

~object_detection_voxel_size (double, default: 0.003)

• The size of the bins when downsampling potential object points before clustering.

~min_object_cluster_size, ~max_object_cluster_size (int, default: 20, 100)

• The minimum and maximum number of points in a cluster for it to be considered an object. The maximum size is there to prevent large objects from being considered, as the robot will not be able to pick these up anyway.

~object_cluster_distance (double, default: 0.03)

• The threshold for clustering when trying to find object clusters.

Pose Generation

~table_edge_padding (double, default: 0.4)

• The amount by which the hull of a detected table is expanded outwards.

~pose_resolution (double, default: 0.05)

• The distance along the expanded hull between subsequent manipulation poses that will be generated.

~robot_side_padding (double, default: 0.2)

• For a manipulation pose to be considered reachable, poses that are +/- robot_side_padding away from the pose must be found reachable by the navigation stack. This provides an extra level of safety to prevent the choosing of poses that are considered reachable but the robot would have a hard time actually getting there.

~edge_angle_neighbor (int, default: 2)

• The edge angle of the expanded hull at each pose point is computed as the angle formed by that pose point plus the x-th neighbor before and after the pose point, where x is edge_angle_neighbor. Larger values of edge_angle_neighbor minimize the effect of a "jagged" table hull, but also has the effect of smoothing table corners.

~max_planner_tries (int, default: 5)

• The maximum number of times the navigation stack planner will be called to compute the reachability of a pose (in case the service call return an error) before marking that pose as unreachable. Needed when robot loads are high and the planner has a hard time responding 100% of the time.

~object_distance_max (double, default: 1.1)

• The maximum distance an object can be from a pose point and be considered reachable by the robot.

Required Services

multi_table_node requires an external service to test the reachability of candidate poses, which is usually done with the navigation stack. In a launch file, the topic name below should be remapped to the appropriate service, such as the one given.

nav_plan_service (nav_msgs/GetPlan)

• The navigation plan service. On the PR2, typically /move_base_node/make_plan

Subscribed Topics

cloud_new_in (sensor_msgs/PointCloud2)

• The point cloud that will be processed for table and object detection.

Published Topics

markers_out (visualization_msgs/Marker)

• Various markers that show the state of table and object detection.

Provided Services

multi_table_detection (multi_table_detector/MultiTableDetection)

• Service call to initiate the multi table detection. The request is empty and the response contains details on the tables detected, as well as the final status of the detection.

Example Usage

Before launching the multi_table_node, you should make sure to correctly configure the source of the point clouds that will be processed. Although any point cloud source could work, the best results are obtained when a persistent collision map, such as collision_octomap, is used as the point cloud source.

Example Launch File

<launch>

  <node pkg="point_cloud_converter" type="point_cloud_converter" name="point_cloud_converter" respawn="false" output="screen">
    <remap from="points_in" to="collision_point_cloud" />
    <remap from="points2_out" to="collision_point_cloud2" />
  </node>

  <node pkg="multi_table_detector" name="multi_table_node" type="multi_table_node" respawn="false" output="screen">

    <!--topic remapping-->
    <remap from="cloud_new_in" to="collision_point_cloud2" />
    <remap from="markers_out" to="tabletop_detector_markers" />
    <remap from="nav_plan_service" to="move_base_node/make_plan" />

    <param name="up_direction" value="1.0" />
    <param name="table_detection_voxel_size" value="0.04" />
    <param name="object_detection_voxel_size" value="0.02" />
    <param name="min_table_cluster_size" value="200" />
    <param name="min_object_cluster_size" value="10" />
    <param name="object_cluster_distance" value="0.04" />
    <param name="table_z_filter_min" value="0.6" />
    <param name="table_z_filter_max" value="1.2" />
    <param name="object_z_filter_min" value="0.03" />
    <param name="object_z_filter_max" value="0.5" />
    <param name="object_distance_max" value="0.9" />
    <param name="table_edge_padding" value="0.55" />
    <param name="pose_resolution" value="0.3" />
    <param name="robot_side_padding" value="0.25" />    
    <param name="edge_angle_neighbor" value="3" />
        
  </node>

</launch>