deno750 / homework_1

Homework 1 - Perception

Homework 1 - Perception

Task 1

Suppose to have N=16 objects identified by N different APRILTAG markers (id = 0, ..., 15):

id	frame_id
0	red_cube_0
1	red_cube_1
2	red_cube_2
3	red_cube_3
4	yellow_cyl_0
5	yellow_cyl_1
6	green_prism_0
7	green_ prism_1
8	green_ prism_2
9	blue_cube_0
10	blue_cube_1
11	blue_cube_2
12	blue_cube_3
13	red_ prism_0
14	red_prism_1
15	red_prism_2

• Only n ≤ 16 objects will be randomly placed on the table in front of the robot;
• From command line, a human operator will ask for the recognition of x ≤ n of them. The request will be made through frame_id:

$ rosrun your_package your_node frame_id_1 frame_id_2 ...

Thus, implemente a NODE able to:

• Match the id of an object with its frame_id;
• Accept inputs from command line;
• Read a /topic publishing the pose of each requested object;
• Store these poses on a text file.

Note:

• The APRILTAG library detects the markers through their ids;
• The library outputs the pose (position and orientation) of each marker.

Apriltags example:

• Open a shell and launch the simulated environment in Gazebo:

$ roslaunch robin_arena robin_bringup.launch simulation:=true

• Place some objects on the table in front of the robot and, on a new shell, launch the AprilTag node:

$ roslaunch robin_arena apriltag.launch simulation:=true

• Open Rviz (third shell):

$ rviz

and ADD the topic /tag_detections_image (You can visualize the published information also by command line: rostopic echo /tag_detections – or launching rqt_image_view and selecting the /tag_detections_image topic)

---

Task 2

As before, n objects are on the table (yellow cylinders, red and blue cubes, red and green prisms). From command line, a user ask for the detection of x ≤ n of them, naming their categories:

$ rosrun your_package your_node red_cube red_cube yellow_cylinder ...

Thus, implement a NODE that exploits the Point Cloud Library to detect each object. The adopted algorithm should exploit colors or shapes in order to recognize objects. E.g., one known PCL 3D object recognition technique is based on Correspondence Grouping

Note:

• Object meshes are provided (in robin_arena/meshes)