We are happy to announce the final results of the 2018 Agile Robotics for Industrial Automation Competition (ARIAC), hosted by the National Institute of Standards and Technology (NIST).
Now in its second year, ARIAC is a simulation-based competition designed to address a critical limitation of robots used in industrial environments: that they are not as agile as they need to be. Many robots are not able to quickly detect failures, or recover from those failures. They aren’t able to sense changes in their environment and modify their actions accordingly. The goal of ARIAC is to enable industrial robots on workshop floors to be more productive, more autonomous, and more responsive to the needs of shop floor workers by utilizing the latest advances in artificial intelligence and robot planning.
While autonomously completing order fulfillment tasks, teams were presented with various agility challenges developed based on input from industry representatives. These challenges include:
Failing suction grippers, requiring teams to determine if products dropped from the gripper should be retrieved or re-positioned,
Reception of updated/high-priority orders, prompting teams to decide whether or not to reuse existing in-progress shipments being filled or to start new ones from scratch,
Notification of faulty products, requiring teams to replace inadequate products and plan ahead to ensure enough non-faulty products are available for the high priority orders,
Products requested flipped from their original positioning, requiring teams to complete a two-step process to place the product, and
Failing sensors, requiring teams to have a high-level model of the environment to continue working through a complete sensor “blackout.”
Teams had control over their system’s suite of sensors positioned throughout the workcell, made up of laser scanners, depth cameras, intelligent vision sensors, quality control sensors and interruptible photoelectric break-beams. Each team chose a unique sensor configuration with varying associated costs and impact on the team’s strategy. Teams that focused on sensors requiring a higher level of processing -- for example, depth cameras in place of intelligent vision sensors -- gained a points boost for their overall lower system cost. The effect of this aspect of the competition was in full swing in the Finals, with the top two teams choosing sensor configurations that had only a single sensor in common.
The virtual nature of the competition enabled participation of teams affiliated with companies and research institutions from a range of countries. The diversity in the teams’ strategies to solving the agility challenges can be seen in the video of highlights from the Finals:
Scoring was performed based on a combination of efficiency, performance and cost metrics over 15 trials. Additionally, judges awarded points for novel, industry-implementable approaches to solving the agility challenges. The overall standings of the finalist teams are as follows.
First place: Sirius. Dan Barry, Denbar Robotics.
Second place: Pajamas. Joey Gannon, Pittsburgh, PA.
Third place: AAU. Robotics, Vision and Machine Intelligence lab at Aalborg University Copenhagen.
Fourth place: Pack Swiftly. Steven Gray.
Fifth place: Team CASE. Department of Electrical Engineering and Computer Science at Case Western Reserve University.
Sixth place: Virsli Team.
The top three eligible teams will receive cash prizes.
Top-performing teams will be invited to present at an upcoming workshop that will be open to all, including those that did not participate in ARIAC. In addition to showcasing the various approaches used in the competition, we will also be exploring plans for future competitions. If you are interested in giving a presentation about agility challenges you would like to see in future competitions, please contact Craig Schlenoff (firstname.lastname@example.org).
Congratulations to all teams that participated in the competition!