Magazine article AI Magazine

Amazon Picking Challenge 2015

Magazine article AI Magazine

Amazon Picking Challenge 2015

Article excerpt

The first Amazon Picking Challenge (APC) was held at the 2015 International Conference on Robotics and Automation in Seattle Washington, May 26-27. The APC follows in the footsteps of other robotics competitions, most notably Rob°Cup and the DARPA challenges, by posing a real-world environment in which robots must perform humanlike tasks. The APC's focus is on one core - but extremely important - area of robotic competency: manipulating objects in the world.

The competition scenario was a Kivalike warehouse in which the robot had 20 minutes to pick items off a shelf and put them into a plastic tote. The 12 bins on the shelf were stocked with 25 products that posed a range of perception or manipulation challenges. Each bin had one target item. A robot received a base score of 10 points for successfully picking the target item, with bonus points for cluttered bins or difficult items. Participants could lose points by picking a wrong item, damaging, or dropping items (figure 1).

Amazon organized the competition and provided $26,000 USD in prizes and $70,000 USD in travel grants. Several competitors also had technical or financial support from hardware vendors. Twenty-six teams, representing 11 different countries, made the trip to Seattle. A wide variety of robots were represented, including most of the commercial vendors and several custom-made devices. An abundance of end effectors were exhibited, including grippers, pincers, spatulas, and suction cups.

The competition was won by RBO from the Technical University of Berlin who picked 10 items and scored 148 points. MIT placed second with 7 items generating 88 points. Third place was Team Grizzly from Dataspeed Inc. and Oakland University with 35 points. Many other teams looked promising during the setup period but for various reasons failed in their official 20-minute trial. The causes of failure included last minute code changes, problems adjusting to the exhibit hall lighting, and modeling errors. While most of the robots were stationary, it is interesting that two of the top three were mobile.

In the first APC, the majority of the teams focused on traditional perception and robotics problems: identifying an object, figuring out how to grasp it, and then computing joint motions to accomplish the desired trajectories. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.