Artificial Intelligence Wiki

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.
====== Homework: ROS packages for perception ======

The aim is to create ros packages and set up a small pipeline for finding objects on top of a table offering a service for returning the position and the volume of these. For the sake of simplicity we assume that these objects are clearly separated, meaning that they are spread around the table in an orderly manner, neiter  touching nor occluding each other.

{{ :teaching:se-kiba:ros-homework-perception.png?nolink&700 |}}

===== Getting started =====

Please install the system according to the instructions here: https://ai.uni-bremen.de/wiki/software/ros/installation

Before you start workin on the assignment, make sure to have completed the 'beginner level' tutorials from [[http://www.ros.org/wiki/ROS/Tutorials]] and you understand how to **subscribe to a topic and create a new service and message types.**

===== First steps =====
We have created a git repository for the code you will develop at:
  https://github.com/ai-seminar/group-perception
Add this to your ROS workspace:  
  rosws set group_perception --git https://github.com/ai-seminar/group-perception.git
  rosws update
  source ~/.bashrc
  
Create the three ROS packages and as a first step commit and push the empty packages to the repository.

Download the bag file from this [[https://ai.uni-bremen.de/public/kiba_pointcloud.bag|link]]. 
====== Package Descriptions ======

===== 1. perception_pkg =====

This package is the core of the pipeline. You need to implement three main parts here. First a subscriber to the topic where point clouds are being advertised, then a processing module, where you work on the point cloud that you read in from the topic, and lastly the same node needs to offer a service for returning results. 

====Subscribe to the topic====

As shown in the tutorial. Example code is [[https://github.com/ai-seminar/perception-tutorials|here]]

====Processing the cloud====

Point Clouds should be processed as they are read in, and a buffer of the results should be kept in memory at all time. The processing part of the package is to segment out objects lying on a table and separate each object in a separate cluster. After having done so, you are required to calculate the centroid of each cluster as well as the volume of it (note that all values in PCL are represented in meters.), and store these in a buffer. You will need to advertise a service, which, upon request, will always return the latest results. 

Notes on processing the input clouds:
  * Start from the code presented in the tutorial ([[teaching:se-kiba:perception-tutorial|Perception Tutorial]])
  * You are allowed to filter the cloud on the z axis only between 0 and 1.5 meters
  * during plane fitting using RANSAC, make sure you find the correct plane.
  * in order to find the objects on the table create a Convex Hull and use ExtractPolygonalPrism in order to find the points above the plane.(more on this [[http://pointclouds.org/documentation/tutorials/hull_2d.php#hull-2d|here]] and [[http://docs.pointclouds.org/trunk/classpcl_1_1_extract_polygonal_prism_data.html|here]]) 
  * use Euclidean clustering with a distance threshold of 3 cm-s to find the point cluster of each object [[http://www.pointclouds.org/documentation/tutorials/cluster_extraction.php|link]]
====Service for advertising the result====

Results should be returned in the message type described in the next section. Upon service call the latest results need be returned. Results are in the form of a list containing //(c_centroid, c_volume)// pairs, beginning with the largest in a descending order.

===== 2. perception_group_msgs =====

This package should contain all the message and service type definitions used by the //perception_// and //perception_client_// packages. 
You should define a new message type. This is the type that the service, offered by the perception package, will be returning. Call your new message type **PerceivedObject**. It needs to contain a unique ID, a center point, and the volume of the object. Remember to place the defined message in the **msg** folder of the current ros pkg, and uncomment the necessary lines in the cmake file, so the necessary files get generated upon rosmake.
Note:Center pose needs to be a //Point//, which is defined in the //geometry_msgs// package.

You will also need to define a new service in this same package. This service, call it **GetClusters**, will be able to send a request in the form of a string to the package providing it and its response is going to be a list of **PerceivedObject**-s


===== 3. perception_client_pkg =====

This package implements the client side of the pipeline. It is responsible for querying the service for the latest results, and for communicating them to the users. Besides serving as the bridge between users and processing, it is also the entry point for the pipeline, meaning that it needs to offer an easy interface for launching everything that is necessary for the pipeline (roscore, openni, service subscriber etc. )
 

For questions and clarifications contact //balintbe **at** tzi **dot** de//