The node then proceeds to grasp the object by sending our goal joint positions.
## Results and Expectations
This demo serves as an experimental setup to explore object grasping with Stretch. Please be advised that this code is not expected to work perfectly. Some of the shortcomings of the demo include:
- The grasp object candidate detection often fails due to the quality of the input point cloud.
- After grasp candidate detection, the segmented point cloud might not cover the entire surface for an object. Hence, the calculated grasp point will have a constant offset from the ideal grasp location.
Users are encouraged to try this demo and submit improvements.
This demo serves as an experimental setup to explore object delivery with Stretch. Please be advised that this code is not expected to work perfectly. Some of the shortcomings of the demo include:
- The node requires the target user's face to be in the camera view while triggering the demo. As it stands, it does not keep any past face detections in its memory.
- Facial landmarks detection might not work well for some faces and is highly variable to the deviation from the faces that the algorithm was originally trained on.
Users are encouraged to try this demo and submit improvements.
This demo serves as an experimental setup to explore whiteboard writing with Stretch. Please be advised that this code is not expected to work perfectly. Some of the shortcomings of the demo include:
- Each Stretch has unique calibration thresholds. The contacts thresholds defined in the stock demo code might not work for your Stretch. Additional tuning might be necessary.
- The quality of the written text structure is also highly dependent on mobile base movements. Currently, navigation is open-loop and does not account for accumulated error terms.
Users are encouraged to try this demo and submit improvements.
This demo serves as an experimental setup to explore drawer opening with Stretch. Please be advised that this code is not expected to work perfectly. Some of the shortcomings of the demo include:
- The demo performs a blind hook operation using contacts sensitivity. Special care must be taken so as to not damage the target surface.
- The code might command the Stretch to lift its arm by a large amount which might cause collisions in the workspace. The code also does not take into account the configuration space of the robot.
Users are encouraged to try this demo and submit improvements.
@ -103,3 +103,13 @@ if True and (strokes is not None) and (len(strokes) > 0):
```
After validation, the node then proceeds to execute the wiping plan in a sequential manner.
## Results and Expectations
This demo serves as an experimental setup to explore surface cleaning with Stretch. Please be advised that this code is not expected to work perfectly. Some of the shortcomings of the demo include:
- The area of surface being cleaned is dependent on the fill rate of the RealSense point cloud. If the segemented cleaning surface has too many holes, Stretch might only attempt to clean a fraction of the total area.
- As each Stretch is unique in its contacts thresholds after calibration, your robot might attempt too aggressively to press down the cloth upon the surface. Additonal tuning might be necessary.
Users are encouraged to try this demo and submit improvements.