Team RuBot’s experiences and lessons from the ARIAC☆
Introduction
Millions of industrial robots are currently employed in modern factories that significantly improve productivity and reduce repetitive manual labor. In a vast number of industrial robotics applications, the pick-and-place operation plays an indispensable role in moving parts and products.
To help enhance the agility of industrial robots, the National Institute of Standard and Technology (NIST) initiated the Agile Robotics for Industrial Automation Competition (ARIAC) in 2017 [1] and started to give cash awards to top 3 winners one year later. The competition focuses on testing the agility of industrial robot with pick-and-place tasks and has been held for four consecutive years.
Each year’s ARIAC comprises of at least two rounds, including one or more qualification rounds, where every eligible person or team can register at the beginning, and a final round among 5 teams selected based on their performance in previous rounds. The competition has seen an increasing number of registrations. In 2020, over 100 teams registered in the qualification round.
ARIAC is a simulation-based competition using Gazebo [2], an open-source 3D simulator by Open Source Robotics Foundation (OSRF). Gazebo Environment for Agile Robotics (GEAR), designed by OSRF and maintained by NIST, is used as an interface software between competitor’s package and the competition environment. In the competition, participants are required to design a system in ROS [3] to fulfill a set of shipments. One or more robots are provided, and participants are to decide among a variety of sensor choices including break beam, proximity sensor, laser scanner, camera, and so on, to deploy in the environment. The anticipated system should be able to pick products from a set of places including bins, shelves, and a conveyor belt. The products are then to be placed on the tray of an Autonomous Guided Vehicle (AGV) to fulfill a shipment. Scoring metric of ARIAC takes into account the number of products successfully shipped, the pose of the shipped product, the time used for each shipment, and the costs of sensor deployment. The contest aims to simulate the real industrial environment for pick-and-place to the greatest possible extent by considering an exhaustive list of events that could occur in a warehouse environment and selecting a set of agility tasks to fulfill. So far, organizers of ARIAC have selected eight agility challenges based on industry feedback and brought them into the competition scenarios. These eight scenarios are: presence of faulty products, insufficiency of products, flipped products, in-process order update, sensor blackout, dropped products, in-process high priority order interruption, and moving obstacles.
Rutgers Algorithmic Robotics and Control Lab (ARCL) started competing in ARIAC since 2019 with its Team RuBot entry, consisting mainly of Ph.D. students. Team RuBot placed 2nd in ARIAC 2019 and 3rd in ARIAC 2020. In this article, we share our journey through the competitions and our lessons learned in navigating the challenging multi-objective optimization tasks. To that end, we start with a brief overview of ARIAC and related research. We then outline our system design for ARIAC 2019 and ARIAC 2020. As these sketches are unfolded, we discuss the challenges we faced and how we addressed them.
Section snippets
Background
Since the inception of ARIAC, the essential task for the required robotic systems has been picking a set of products (pulley, disk, piston rod, gear and gasket) from a conveyor belt, bins, or shelves and placing them on a tray for delivery. In addition to the main pick-and-place task, eight agility challenges, listed below, were provided to test the agility of the systems designed by competition participants.
- 1.
Faulty Products. Some products are faulty. After a faulty product is placed on the AGV
System overview
Competitors in ARIAC are tasked to design a software in the form of an ROS package and provide necessary scripts for installing dependencies and running the software. The software must be capable of retrieving the sensing data and control the robots in Gazebo simulations within an Ubuntu Bionic and ROS Melodic environment. Team RuBot’s submissions were implemented in C++ for both ARIAC 2019 and ARIAC 2020. No auxiliary software dependencies exist in our robot control sub-system.
Team RuBot takes
Learning from failures
In 2019 and 2020 ARIAC final rounds, 15 tests covering all the agility challenges were performed on the competitors’ software. From the logs traces of the finals tests, we could identify where the systems failed. As a matter of fact, our systems are still far from being perfect. In 2019, 4 out of 15 tests were not successfully carried out, while the number of failed runs increased to 5 in 2020.
Compared with other mistakes like lacking precision for products’ poses on the tray or missing
Conclusion
Considering the amount of effort we put in, our team excelled in producing an agile and well-functioning system. Our submissions proved to be competitive, placing among the top three in both 2019 and 2020. We have summarized the approaches we took, including our task and motion planning architecture, pose representations, and the choices of sensors. Our system’s failure modes, which we learned the hard way, suggest that it is important for systems to be able to agilely recover from errors and
CRediT authorship contribution statement
Si Wei Feng: Methodology, Software, Investigation, Writing - original draft, Visualization. Teng Guo: Software, Investigation, Writing - review & editing. Kostas E. Bekris: Writing - review & editing, Supervision. Jingjin Yu: Writing - review & editing, Resources, Supervision, Funding acquisition.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgment
The authors thank the organizers of the ARIAC for providing opportunities, scoring and feedbacks for the our submissions: Anthony Downs, Dr. William Harrison, Dr. Zeid Kootbally, Dr. Craig Schelenoff from NIST, and Shane Lorentz from OSRF as well as the expert judges for the final round. This work is supported by National Science Foundation, United States of America awards IIS-1734419 and IIS-1845888.
References (27)
- et al.
Automated material handling in composite manufacturing using pick-and-place systems – a review
Robot. Comput.-Integr. Manuf.
(2018) - et al.
Parallelism of pick-and-place operations by multi-gripper robotic arms
Robot. Comput.-Integr. Manuf.
(2016) - et al.
Addressing perception uncertainty induced failure modes in robotic bin-picking
Robot. Comput.-Integr. Manuf.
(2016) - et al.
Robot programming using augmented reality: An interactive method for planning collision-free paths
Robot. Comput.-Integr. Manuf.
(2009) - et al.
The agile robotics for industrial automation competition
AI Mag.
(2018) - et al.
The player/stage project: Tools for multi-robot and distributed sensor systems
- M. Quigley, K. Conley, B. Gerkey, J. Faust, T. Foote, J. Leibs, R. Wheeler, A.Y. Ng, ROS: An open-source robot...
- et al.
Moveit!
IEEE Robot. Autom. Mag.
(2012) - . NIST Engineering Laboratory Intelligent Systems Divison, ARIAC rules,...
Planning collision-free motions for pick-and-place operations
Int. J. Robot. Res.
(1983)
Complexity results and fast methods for optimal tabletop rearrangement with overhand grasps
Int. J. Robot. Res.
DIPN: Deep interaction prediction network with application to clutter removal
Task-level planning of pick-and-place robot motions
Computer
Cited by (0)
- ☆
Our submissions to ARIAC are completely open sourced at Github. Team RuBot 2019: https://github.com/ustcsiwei88/RuBot Team RuBot 2020: https://github.com/ustcsiwei88/RuBot2020.