Amazon Picking Challenge (2016)

Project Overview

This project aims to build an autonomous system that is able to effectively able to retrieve and sort items from a set of shelves to a basket/tote, then do the reverse. This task combines aspects from multiple disciplines (computer vision, machine learning, actuator control and manipulation) to make a complete robotic system. The project was invoked by Amazons yearly competition, “The Amazon Picking Challenge”. Although it was created for the challenge, any research in these areas also apply to a broader context in research and industry.

Project Objectives

The objective of making a robot pick and place items off a shelf is no easy feat. Researchers are working on:

  • Ensuring robust detection of the objects on the shelf (which is shiny and reflective).
  • Integrating a reliable pick and place action involving a tight coupling of vision and action.
  • Using Baxter as a platform with a few hardware modifications.
  • Using both RGB and depth cameras (real sense) for robust object detection.

Project Milestones

At the 2016 Amazon Picking Challenge in Germany, 16 teams from around the world competed. The Australian Centre for Robotic Vision (ACRV), headquartered at QUT, reached sixth place in the picking task using a Rethink Robotics Baxter.


Team

Other Team Members

PhD: Adam Tow James Sergeant Peter Jujala Trung Pham Fangyi Zhang QUT Students: Matthew Cooper Jake Dean Lachlan Nicholson Ruben Mangels

A demo of the Baxter robot

Object manipulation

Researchers build simulation programs to ensure that all planning is done in a safe environment