R12 robotic arm has shoulder, elbow, waist and twist joints to move the end effector. A camera is attached to the end effector. The purpose is to find the corners of a box that could be present anywhere in its workspace.
Click here for the project link
Algorithm Methodology
- Robot will take a series of pictures by dividing the workspace into sectors
- Use Computer vision to know about the status of the object in the sector
- Moves to the most appropriate quadrilateral in the sector where the possibility of box is high
- Subsequently takes pictures and moves to a corner depending on custom developed Fuzzy logic
Robot Controller
- Built a Custom GUI to control the Robotic arm remotely and for calibration
- Robot controlled via HTTP protocol
Presented the GUI to future students for using the Robot easily
Image Processing
- The image is blurred, applied threshold based on Hue and found the centroid based on the moments
- This is used to detect the quadrilateral of the image in the sector workspace
- Later, I found the possible contours, applied convex hull to the max size contour and applied an approximate polygon on top of it
Quantization of the image
- The image is quantized into to 3X3 grid based on the presence of the box, where each cell represents the ratio of the box that fills the space
- It can detect edge, corners using that
- Depending on the observation a weighted vector score is computed to make the Robot move for the next iteration.
Tools and Libraries Java, JavaFX, OpenCV, Eclipse
