Dutch Robotics Wizards Win Amazon Warehouse Stocking Challenge With Machine Learning Robot Arm
Human hands have built a great many things, including robots and machines that replace human hands. In celebration of this transition from flesh and blood to metal and lubricant, Amazon's annual Picking Challenge tasks those in the field of robotics to compete in a competition to build automated systems that can simplify the task of picking items from shelves. This year it was won by Team Delft, a collaboration between TU Delft Robotics Institute located in the Netherlands and Delft Robotics.
The challenge is separated into two finals, one for stowing and one for picking. For the stow task, robots had to automatically grab a range of products from a container and put them on shelves. And for the pick task, the robots had to do the exact opposite—take items off of shelves and place them in a container. Team Delft won both finals by collecting 214 points in the stow task and 105 points in the pick task, making them a double champion.
The challenge is separated into two finals, one for stowing and one for picking. For the stow task, robots had to automatically grab a range of products from a container and put them on shelves. And for the pick task, the robots had to do the exact opposite—take items off of shelves and place them in a container. Team Delft won both finals by collecting 214 points in the stow task and 105 points in the pick task, making them a double champion.
Each task in the two finals has a time limit and there's a scoring rubric (PDF) that assigns a point value to successfully moving target items to the correct location. Teams lose points when one of the following occurs:
"We built a very robust system, that hardly makes any mistakes in picking the products, thanks to our expertise in ‘bin picking’. In addition, the robot chooses the maximum of points that can be scored for each product to pick," said Kanter van Deurzen from Delft Robotics.
There were 16 finalists in the competition. Team Delft built a flexible robot system based on industry standards with a robot arm boasting seven degrees of freedom, high-quality 3D cameras, and a gripper the engineers developed in-house.
On the software side, the team integrated advanced code based on state of the art artificial intelligence techniques and robotics. The components were developed with the Robot Operating System for Industry (ROS-Industrial), which will later be released as open software.
- Removing non-target items from the shelf and not replacing them
- Reporting an incorrect final position for an item in Task Output File
- Damaging any item or the shelf
- Dropping an item from a height of more than 0.3 meters
- Leaving an item protruding out of its bin by more than 0.5 centimeters
"We built a very robust system, that hardly makes any mistakes in picking the products, thanks to our expertise in ‘bin picking’. In addition, the robot chooses the maximum of points that can be scored for each product to pick," said Kanter van Deurzen from Delft Robotics.
There were 16 finalists in the competition. Team Delft built a flexible robot system based on industry standards with a robot arm boasting seven degrees of freedom, high-quality 3D cameras, and a gripper the engineers developed in-house.
On the software side, the team integrated advanced code based on state of the art artificial intelligence techniques and robotics. The components were developed with the Robot Operating System for Industry (ROS-Industrial), which will later be released as open software.
