A team of engineers from Delft is participating in the Amazon Picking Challenge at the end of June. During this challenge, robots will retrieve a wide range of products from shelves and put them in a container, and vice versa. According to the Delft team, robustness of the robot will be key to success.

The Amazon Picking Challenge 2016 is part of RoboCup 2016, the famous international robot competition which will be held from 29 June to 3 July in Leipzig. In this challenge, robots equipped with grippers will autonomously retrieve a wide range of products from their shelves and put them in a container (the Pick task), and in a second round retrieve the products from the container and put them back on the shelves (the Stow task).  In order to do this, the robot needs to perform complex tasks such as planning, recognising products and manipulating them.

Unsolved problems in automation

At first glance, this may not seem like a difficult challenge, but the products on the shelves are randomly placed in any orientation. Besides, there is a wide variety of products, ranging from books to T-shirts, with different sizes, shapes, colours. The robot thus needs deal with unsolved problems for automation: handling variety and operating in an unstructured environment. Last year, during the first Amazon Picking Challenge, the winning team was able to successfully collect 11 out of 12 products from the shelves, while the runner-up was only able to collect 7.

The system

Team Delft is one of 16 finalists for the Amazon Picking Challenge, and is a collaboration between TU Delft Robotics Institute and the company Delft Robotics. The team is building a flexible robot system based on industry standards. The system is equipped with a robot arm with seven degrees of freedom, high-quality 3D cameras and an in-house developed gripper. To control the robot, the team is integrating advanced software components based on state of the art artificial intelligent techniques and robotics. The components are developed with the Robot Operating System for industry (ROS-I), and will be released as open software.


The team is led by TU Delft Robotics Institute researcher Carlos Hernandez Corbato and Kanter van Deurzen from Delft Robotics. ‘We are focusing primarily on the robustness of the system,’ says Van Deurzen. ‘That turned out to be the biggest stumbling block for last year’s competitors. This is also why we are relying on industry standards.’

Van Deurzen admits that it will still take a number of years before systems like this are mature enough to be successfully used in the real world. First, the system must be sufficiently reliable and only then can aspects such as speed be improved. ‘Nevertheless, these developments are still very interesting for big companies like Amazon. The degree of automation can thus be increased even further, with humans keeping an eye out on the robots. The collaboration and coordination between robots and humans is certainly a major focus in this field.’


‘One of the special things about the Delft system is the collaboration between a university and a market player,’ Carlos Hernandez Corbato states. ‘We complement each other very well.’ The team is also supported by RoboValley, a unique network consisting of scientists, industry and government agencies who are working on the next generation robotics, with TU Delft Robotics Institute at its heart.

Team Delft conducted the first tests with the system during RoboBusiness Europe in Odense, Denmark (1 to 3 June).

Team Delft is supported by RoboValley and the EU project Factory-in-a-Day.

More information

Website Amazon Picking Challenge 
Blog Team Delft APC 2016
Team Delft APC 2016 on Twitter
Article ‘Team Delft APC 2016 Progress’ on ROS Industrial
Film of Team Delft APC's robot on YouTube  

Contact information

Carlos Hernández Corbato (TU Delft) C.H.Corbato@tudelft.nl
Kanter van Deurzen (Delft Robotics) k.vandeurzen@delftrobotics.com
Ilona van den Brink (communication advisor TU Delft), i.vandenbrink@tudelft.nl +31-15-2784259