Mailman

Sensing

• Camera sensing with AR Tags.
• Closed-loop IR sensing to confirm the gripper is close enough to the target and adjust pose if not.

Planning

• Customized Moveit!(Motion Planning Package) to enhance planning performance.
• Avoid collision objects(table) in planing
• Intermediate poses to encounter the uncertainty of Moveit! pkg.

Algorithm

IMAGE PRE-PROCESSING:

• RGB to Gray Conversion
• Adaptive Thresholding to Remove Noise
• Contour Detection: recognize the boundary of digits and draw rectangle boundary
• Segment and Resize Image(28x28)

FEATURES:

• Background Pixel Range: After resizing the cropped image, we extracted the pixels from the image

CLASSIFICATION:

•  K-nearest neighbors（KNN): supervised machine learning(K=1) MNIST handwritten database Euclidean Distance for the images by using MNIST

For more information: Mailman Baxter

Github Link：Code