Watch the video below to see these inventive teams – and their robots – in action.

Congratulations to this year’s winners from the Australian Centre for Robotic Vision. See the full results here.

RBO Team from TU Berlin wins the Amazon Picking Challenge at ICRA 2015. Photo Credit: RBO.

Capturing and processing camera and sensor data and recognizing various shapes to determine a set of robotic actions is conceptually easy. Yet Amazon challenged the industry to do a selecting and picking task robotically and 28 teams from around the world rose to it.

Perception isn’t just about cameras and sensors. Software has to convert the data and infer as to what it “sees”. In the case of the Amazon Picking Challenge held last week at the IEEE International Conference on Robotics & Automation (ICRA), each team’s robot was to pick from a shopping list of consumer items of varying shapes and sizes – from pencils, to toys, tennis balls, cookies and cereal boxes – which were haphazardly stored on shelves, and then place their selected items in a bin. They could use any robot, mobile or not, and any arm and end-of-arm grasping tool or tools to accomplish the task.

It’s tricky for robots using sensors to identify and locate objects that can be confused by plastic packaging within the shelf or storage area. Rodney Brooks, of iRobot, MIT and Rethink Robotics fame, often speaks of an industry-wide aspirational goal regarding perception in robotics: “If we were only able to provide the visual capabilities of a 2-year old child, robots would quickly get a lot better.” That is what this contest is all about.

Software has to first identify the item to pick and then figure out the best way to grab it and move it out of the storage area. Amazon, with its acquisition of Kiva Systems, has mastered bringing goods to the picker/packer and now wants to automate the remaining process of picking the correct goods from the shelves and placing them in the packing box, hence their Amazon Picking Challenge.

The top three winners were the teams from Technical University (TU) – Berlin, with 148 points; MIT in 2nd place with 88 points; and the 3rd place finisher (Oakland U and Dataspeed), which only got 35 points. Teams were scored on how many items were correctly selected, picked and placed.

Many commercial companies with proprietary software for just this type of application (such as Tenessee-based Universal Robotics and their Neocortex Vision System, and Silicon Valley startup Fetch Robotics who were premiering their new Fetch and Freight system at the same ICRA conference) chose not to enter because the terms of the challenge included that the software be open sourced.

Team RBO from TU-Berlin wrote their own vision system software and will soon be working on a paper on the subject. They used a Barrett WAM arm because it was the most flexible device for this task amongst the arms that they had access to in their lab. They used a vacuum cleaner tool augmented with a suction cup and a vacuum cleaner to power the suction. For a base, they decided they needed to go mobile and used an old Nomadic Technologies platform, which they upgraded to fit the needs of the contest. Nomadic no longer exists. It was acquired by 3Com in 2000. TU’s photo at right shows robot’s camera view and random placement of items in cubby holes. Watch a video of their winning run here:

Noriko Takiguchi, a Japanese reporter for RoboNews.net, who was at the contest, observed the TU team and said that their approach was torque-based plus position control of the arm and the mobile base, consequently they had good torque control that that gave them flexibility in choosing where to place the suction cup and how much suction to apply.

Team RBO received a 1st place prize of $20,000 plus travel costs for the equipment and team members.

The Amazon Picking Challenge is over and two things stood out. One: how many different arm gripper solutions were possible; and two: just how difficult the challenge still is. The gap between the top two teams and the other 26 teams was significant, with Team RBO scoring 148 points, Team MIT scoring 88 points and Team Grizzly next best with 35 points.

Just in: video from winning Team RBO (hattip to SushiCapacitor on Reddit)

https://youtu.be/LtWPH-bcc4M

The Amazon Picking Challenge was developed to spur advancement in fundamental technologies for automated picking in unstructured warehouse environments. This is something that Amazon badly needs to do better in order to fulfill all our e-commerce instant delivery orders. And it’s also something that will enable robots to better work in all areas of our life. Not surprisingly then, there was a lot of excitement at ICRA around the 2 day challenge.

One of the conference expo halls was given over to robot competitions. There were 16 bays with a set of warehouse shelves in each. Amazon had announced a menu of common e-commerce items that would be used in the challenge, but before each round the shelves were stocked with a randomized selection of items from the menu and each team was given a order list to fulfill. Each of the 28 teams was given time slots to practice in and then compete. Over the course of 2 days, 28 teams rotated through the competition.

It was similar to the DARPA Robotics Challenge, at least the 2013 first round. “Like watching paint dry.” With so much happening at ICRA, I was only able to visit the Picking Challenge for short periods of time. As a spectator, I spent most of my time watching robots do nothing. Large amounts of nothing. Occasionally nothing would be enlivened by an attempt to pick up nothing, or perhaps the shelf itself. Once or twice I saw a real pick… get dropped.

If I was very lucky, because the competition area was crowded with media, spectators and members of other teams, I might get to look at the laptops running code and showing representations of what the robot was perceiving while it looked like nothing was happening.

And the perception was the real event. I talked to Team MIT after they posted their score of 88 points, blowing all previous entrants out of the water. I had particularly noticed their combination of an ABB 2 armed robot with a flat scoop and a suction gripper. When asked if they attributed their success to the end effector or other mechanical solutions, the team gave all credit instead to their perception and path planning algorithms.

Ultimately that is the secret sauce that Amazon, and all the other major robot companies, would like to capture; better algorithms for existing robots. The range of hardware in use in the competition was broad. There were several Baxters, Yaskawa Motomans, Universal Arms, ABBs, PR2s, Barrett Arms, custom built 3d printer style rigs and factory automation. There were scoops, hands, grippers, and suction.

End effectors were made or covered in a range of substances, wood, metal, plastic, soft silicon. While it might all be about the algorithm, I will point out that the two most efficient teams did both use suction and soft surfaced grippers. A third thing that stood out to me was how many large robotics companies were in the audience of the Picking Challenge, watching.

Congratulations to the winning teams – and to Amazon for the support they gave the robotics community for this event. Amazon awarded travel grants to ICRA, arranged practice equipment and a $26,000 prize pool.  Participants will be encouraged to share and disseminate their approach to improve future challenge results and industrial implementations.

Some more information from Barrett Technology:

Of 25 teams from around the world, the winner of the Amazon Robotic Bin-Picking Challenge is the Technische Universität Berlin using Barrett’s WAM robotic arm.

The Amazon Picking Challenge is the centerpiece of the record-attendance IEEE Conference on Robotics & Automation in Seattle this year.  There are 28 teams here from around the globe who have brought their hardware and software to the competition.  Entries include robotic arms from ABB, Fanuc, Rethink Robotics, Universal Robots, and Yaskawa-Motoman.  The competition just finished, and the scores are in…  Technische Universität Berlin’s Robotics and Biology Laboratory using Barrett’s WAM arm in 1st place at 148 points followed, in 2nd place with 88 points, by the Massachusetts Institute of Technology using an ABB arm and a gripper with creative finger geometries.  The 3rd place finisher came in at 35 points.

Prof. Oliver Brock of TU-Berlin remarks: “This has been a fantastic team effort.  Every single member of our team contributed with enthusiasm and ingenuity, enabling us to produce a compelling showcase for mobile manipulation as a winning approach to industrial manipulation.”  Speaking to media, Townsend adds: “We are grateful that TU-Berlin won this competition, and of course especially grateful that they chose to use the WAM arm.  The WAM’s dexterity is world-class, the long-and-slender links allow the robot to reach easily in and around tight workspaces, and the WAM’s famed backdrivability enables deft interactions with the environment. While it’s a powerful combination of capabilities, the great creativity and dedication of the students and staff under Oliver’s leadership has been essential, and it has been an absolute pleasure supporting them.”

The WAM arm (shown with a BarrettHand).

To learn more about the team from the Technische Universität Berlin, go to

http://www.robotics.tu-berlin.de

Contact Prof. Alberto Rodriguez for more information about the MIT team, at

http://www.cs.cmu.edu/~albertor

To read more about the 2015 Amazon Robotic Bin-Picking Challenge, go to

http://amazonpickingchallenge.org

For information on the IEEE International Conference on Robotics & Automation, go to

http://icra2015.org