In 2014, Amazon sponsored a contest to solve the warehouse order-fulfillment problem of picking a wide range of items from a shelf under uncertain conditions – a task that, in 2014, could only be done by human pickers and one that Amazon would like to automate to better fulfill their increasing load of e-commerce orders.
Terms of the challenge limited the teams to those from academia but it was a fun team-effort well supported by robot manufacturers and the contest was held at ICRA, IEEE’s premier scientific get-together for roboticists. Amazon provided a variety of items to pick covering all sizes, shapes and packaging – including clear plastic baggies – and variable positioning within the shelves. All the major robot companies were present and were looking for better vision and grasping methods – better algorithms – for their existing robots… and that they got. But the speeds were still not ready for prime time.
The 2014 Challenge went v…e…r…y slowly. In 2015, things moved similarly slowly. In 2016, a joint team from the TU Delft Robotics Institute and Delft Robotics won the contest at a speed of around 100 picks an hours with a failure rate of 17% which is 1/4 the speed of a human picker.
The 2017 Challenge was won by a team from Australia, with German and Singaporean teams coming in 2nd and 3rd. The winning teams won $80,000, $40,000 and $20,000 respectively. 16 teams from 10 countries participated. Picking speeds ranged up to 272 while stowing speeds went from 25 to 160 – still far from the average human rate of 400 per hour.
The Australian team from The Australian Centre for Robotic Vision developed their own Cartesian robot, which moves along 3 axes at right angles to each other, and a rotating gripper that enabled picking items using either suction or a two-finger gripper.
Fifteen members of the Centre’s 27-person team of researchers from QUT, The University of Adelaide and the Australian National University, were in Japan at ICRA for the event.