Grasp object demo not working anymore

Ask
1

I cannot get the grasp object demo to try grasping anything anymore. After it says find the floor, it seems it does not get a grasp target and just skips to printing the completion message. The base and arm do not move at all. Does anyone know what could be happening? I have it set up with an object on a clean surface looking at it out over its arm, as it shows for all the autonomy demos. This had been working before and I have not changed the code. Please let me know if anyone has any ideas why it is not able to identify a target object. Thank you!

terminalout
terminalout1199×525 115 KB

grasp_target_20230202083907

2

Hey @jgangemi, thank you for posting this query on the forum. As a quick sanity check, is it possible for you to attach some pictures of your setup? I want to confirm two things:

  1. If the object is being occluded by the gripper.
  2. If the object is within reachable distance of the robot for this demo.

Looking at the second picture, I also suspect that you might need to recalibrate the robot URDF as described in this guide:

Best,
Chintan

3

Here is my set up…

image
image1269×1654 440 KB

You were correct, I needed to recalibrate the robot URDF… I followed the instructions you posted, and the grasp object demo works again as expected! Thank you so much!!

1 Like
4

Brilliant! I am glad that the demo works for you again.

A few notes looking at your setup to avoid future issues:

  1. Since the camera is sensitive to lighting conditions, try to experiment with diffused lighting. For example, I have noticed that a white table like yours appears shiny to the camera with direct overhead lighting or the windows completely open. I noticed a better success rate for my setup when I allowed natural light to pass through blinds that acted as a diffuser.
  2. The object you want to detect must be noticeably higher than the table surface. The grasp object demo uses a Maximum Height Image (MHI) for the segmentation of objects according to their height in the scene. This means that an object which rises higher above the table surface will be easier to detect and have a better success rate than a flatter one.
  3. The demo is prone to errors where the camera detects the arm itself as an object to grasp. To correct this, ensure that the gripper is out of the camera’s view and the camera only sees the object to be grasped in its frame.

All the best and feel free to reach out if you have questions!

Best,
Chintan

1 Like
5

Thank you very much for those tips, I will keep them in mind!

I also spoke too soon about the demo… it is no longer working again. I keep getting the following errors:

WARNING: There are too few points from the depth image for plane fitting, so only using the RGB ArUco estimate. number of points = 0

and the terminal output below

image
image1920×290 62.1 KB

The demo does not even try to initiate. I believe there is something wrong with the camera? Im not sure if you have seen this before?

Thank you again for all of your help with this!

6

@jgangemi I’m not sure what’s going on in there. Could you copy and paste the text from your terminal leading up to the error in a file and email it to support?

Best,
Chintan

1 Like
7

Hey @jgangemi. Changes were pushed to the stretch_ros repository recently. Be sure to periodically pull these changes by executing the following command in the ~/catkin_ws/src/stretch_ros directory:

git pull

and then rebuild the workspace by executing the following command in the ~/catkin_ws directory:

catkin_make

Best,
Chintan

1 Like