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 integrated 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-Industrial), and will be released as open software.
The Challenge consists of two separate finals: a ‘stow task’ and a ‘pick task’. In the stow task, 12 different items of different shapes and sizes, representative of some of the retailer’s bestselling products, were put in a red plastic box. The robots had to pick them out and put each one at a predetermined place on a shelf.
In the pick task, a dozen specific items had to be lifted off shelves containing a mix of goods and packed in the boxes. In some cases, other products had been deliberately placed in the way of the targets.
In both cases, the teams were only given a computer file describing the range of objects involved and instructions for which should be moved, five minutes before they had to start. Once the task began, the robots had to act autonomously. Points were deducted for: damaging any item; dropping the products from a height of more than 30cm; or leaving an object protruding more than 0.5cm outside where it belonged on a shelf.
In the stow task, the Dutch team came close to a perfect score by stowing 11 of the 12 items into the container. More errors were made by all teams in the pick task, which Amazon said was significantly more difficult than in previous years. Team Delft tied on points with Japan's Team PFN, but Delft was given the advantage because it took 37 seconds less time to make its first pick.
Kanter van Deurzen from Delft Robotics said: “We built a very robust system, that hardly makes any mistakes in picking the products, thanks to our expertise in ‘bin picking’. The robot chooses the maximum points that can be scored for each product to pick.”
