Pose Estimation and using it with Navigation Stack in ROS

[Abhijit03]

Hi Construct Community, @marco.nc.arruda @bayodesegun @duckfrost2 @albertoezquerro @rtellez

I am working on a project where I have to estimate the Pose of an object (specifically dustbin) in real time and provide necessary commands to move_base so that I can orient my robot according to it . The CAD model of the object is available to me. I need help from anybody that can kindly guide me on how I can implement this. I urgently need your help and I hope the community at Construct who have contributed to my learning can also guide me on this.

I tried find_2d_object, was not convinced. Please guide me
Regards,
Abhi

[Joseph1001]

Hi @Abhijit03 ,
I am assuming that you want to get the location of a dustin model from Gazebo. If this is the case, the below mentioned steps should help.

Use the service provided by Gazebo to get the location of a model in Gazebo

rosservice call /gazebo/get_model_state "model_name: 'turtlebot3_burger'"

rosservice call /gazebo/get_model_state "model_name: 'Replace_with_your_models_name'"

Now, say you dont know know exactly the name of the model , because it could be Dustbin or Dust_bin or Dustbun2 etc. So inorder to find the actual name of the model, we use a service provided by gazebo

rosservice call /gazebo/get_world_properties "{}"

This gives the result, which includes the names of the models:

sim_time: 366.918
model_names:
  - ground_plane
  - ring_road_v2
  - working_wall
  - working_wall_0
  - working_wall_1
  - working_wall_2
  - triangular_prism_stop1
  - triangular_prism_direccion_obligada1
  - triangular_prism_direccion_obligada2
  - triangular_prism_stop2
  - triangular_prism_pedestrian0
  - triangular_prism_pedestrian1
  - overlord_camera
  - turtlebot3_burger
rendering_enabled: True
success: True
status_message: "GetWorldProperties: got properties"

If this was not the solution you were looking for, you can reply here and we can work from there on.
Best,
-Joseph

[Abhijit03]

Hi @Joseph1001 ,

Really appreciate your reply , I am not working with the gazebo simulation of the environment rather working with real environment and what I want to do is estimate the pose (position and orientation) of the Dustbin placed in the scene using RGB-D camera and feed this pose to the navigation stack as the goal so that the robot can orient/position itself infront of it. The detection of the Dustbin has been already implemented using YOLO

Just to make it more clear , there is this video (Robot Mapping using Laser and Kinect-like Sensor - YouTube) which I found on the the internet that uses find-object-2d package which is exactly what I am trying to do just with a small addition that the robot use the pose to position itself, much like a robot trying to dock for charging but without the use of aruco markers or qr-code.

I tried to work with the (find_object_2d - ROS Wiki) package which exactly fits my requirements but I find it hard to fine tune the parameters and the detector doesn’t updates the detection and pose estimation as I move the camera around the object and I tried to find more information on how to tune the parameters on the GUI but cant find much help (Find-Object).

I would really appreciate any help. Thanks in advance.

Regards,
Abhi

[Joseph1001]

Hi @Abhijit03 ,I am not to figure it out at the moment. Have you tried the 2 courses from Construct under the title perception?

If you haven’t, i highly suggest you do since it has content similar to what you are expecting for your project.