UR5 Parcel Sorter – CSE Senior Design

Team logo

Team Name

Arm Solutions

Timeline

Fall 2020 – Spring 2021

Students

Gregory Ferguson
Krishna Patel
Donny Pham
Inara Rupani
Mathew Zinke

Sponsor

James Staud

Abstract

Using the UR5 robotic arm, our project has been able to successfully identify incoming packages using computer vision. Once a package is identified, the robotic arm uses a suction gripper to attach to the package, pick it up, and place it into another area. A sensor is used in addition to ensure that the gripper is attached. This process allows for automated package movement within a shipping facility.

Background

In the light of coronavirus and the Industry 4.0 revolution, organizations like USPS, FedEx, and UPS have been overwhelmed with parcels and mail. Workers currently spend all hours of the day picking small parcels from a heap and placing them onto a conveyor for processing. Handling these parcels exposes them to germs on the parcels themselves. Our solution is to not only eliminate the risks to the employees but automate and improve the efficiency of thousands of jobs.

Project Requirements

The UR5 shall be able to handle any package weighing at most 11 pounds
The UR5 shall identify, pickup, and place a package using a suction gripper
The system shall identify box/package picking points
The controller shall run using the ROS library
The system shall run on a Raspberry Pi powered by a Linux kernel distro
The system parts shall communicate using TCP
The vision system shall correctly identify the type of package on the conveyor belt after intensive training the data set
A convoluted Neural Network will be used to train and test data sets for efficient image processing of object detection
The server and client-side will run synchronously
A clean workspace will be used to carry out the system functionalities

System Overview

Results

The goal of this project was to create a parcel sorter machine. Due to time constraints and domain knowledge we were able to create a prototype where we integrated a vision system using YOLOv5, detected where the parcel existed on the workbench using arcuro markers and programmed the dynamically based on where the parcel is located. At this point in time system is able to scan the workbench and move the item to the drop location. The sorting of items based on size is not integrated.

Link to Demo Video

Future Work

Eco mode: calculate the most energy efficient kinematics for the robot based on the package size and desired placement
Multiple departure points: have outgoing placement positions to place the package based on shipment information (destination, size, etc)
Upgraded work-cell: implement a conveyor belt system in the work-cell for package delivery
Variable package types: train the AI and integrate sensor for packages of more varying shapes, sizes and mass balances