Winning ARIAC 2020 by KISSing The BEAR: Keeping things simple in Best Effort Agile Robotics

https://doi.org/10.1016/j.rcim.2021.102166Get rights and content

Highlights

  • Best Effort Agile Robotics.

  • High precision vs. Agile robotics.

  • Abnormal is normal.

  • Raised level of abstraction.

  • KPIs and error handling.

Abstract

In this paper we discuss our evolved understanding on the meaning of agile robotics. The lessons learned are concluded from the four years of our participation in ARIAC, the Agile Robotics for Industrial Automation Competition, that we managed to win in 2020. After elaborating on ARIAC tasks, challenges and scoring and their consequences on algorithm design, error handling and software development processes and methodologies, we introduce the concept of Best Effort Agile Robotics (BEAR). The conclusions are mapped to agility standardization directions. The concept of Best Effort Robotics (BER) is also proposed and discussed, highlighting the similarities with best effort networking, a proven concept in Internet technology.

Introduction

In the recent decade there has been an increasing demand from customers towards the manufacturing industry to provide more and more customized products. Personalized production is one of the key motivations for manufacturers to start leveraging new technologies that enable to increase, for instance, the flexibility of production lines. High flexibility in general is needed to realize cost effective and customized production, as well as, easy application development. Fast reconfiguration and agile behavior can be achieved with moving the robot control from the pre-programmed local robot controllers to the cloud. In industrial robotics research, cloud robotics is a major topic, and several studies have recently shown the benefits of connecting robots to a centralized processing entity: usage of more powerful computing resources in a centralized cloud, especially for solving Machine Learning (ML) tasks [1]; lower cost per robot as functionalities are moved to a central cloud [2]; easy integration of external sensor data and easier collaboration or interaction with other robots and machinery [3]; reliability of functions can be improved by running multiple instances as a hot standby in the cloud, and the operation can immediately be taken over from faulty primary function without interruption [4].

In this paper we go through our journey from the participation on the first Agile Robotics for Industrial Automation Competition (ARIAC) organized by the National Institute of Standards and Technology (NIST) in collaboration with IEEE CASE in 2017 to end up winning the competition in 2020 [5]. The competition is a simulation-based contest designed to encourage robot agility research, as well as facilitate technology transfer. During the years we tested and tried various approaches and learnt a lot in agile robot control.

Our learnings with ARIAC led us to introduce the concept of Best Effort Agile Robotics (BEAR). We discuss the similarities of this concept to best effort networking, similarly as some work discussed the analogies of cloud computing and cloud manufacturing [6].

Section snippets

ARIAC history and participation

The general goal of the ARIAC competition series is to motivate further development and adoption of agile industrial robotics by providing an environment where teams could work on solutions towards more productive and autonomous robots that would also require less time from shop floor workers [5]. Agility is defined to address: (a) failure identification and recovery, where robot can detect failures in a manufacturing process and automatically recover from those failures, (b) automated

What makes ARIAC special

In this section we elaborate more on the ARIAC tasks, challenges and scoring, try to identify what makes this agile framework special, and summarize what consequences it has on system performance, error handling, algorithm design, system architecture, and software development processes and methodologies including simulations and testing.

Input to agility standard

Fig. 7 combines the task representation on the main activities including some of the major architectural components of our ARIAC solution, and visualizes the developed methodologies. The figure also shows the mapping of the Agility Standard (see the Project Authorization Request document here [29]) to our software components and tasks. It is clear that ARIAC was designed to challenge most of the topics in some way.

High precision vs. Agile robotics

Expectations on industrial robots can be briefly summarized as follows: high quality, high precision in every sense. Industrial production consists of repetitive tasks that have fully specified workflow. High quality is a must, production must be error free, failure is not tolerated. We accept in exchange that the robots are very expensive, not just the hardware but the complex software part as well. It takes a lot of time to set-up the environment and program the robots. Once the production

Summary and conclusions

In this paper, we discussed our understanding on robot agility, and agile robotic systems as a whole. After discussing the consequences of ARIAC on algorithm design, error handling and software development processes and methodologies, we provided solutions and showed our best practices for the above points, unifying all of that into the concept of Best Effort Agile Robotics (BEAR). The critical aspects and elements were mapped to agility standardization directions.

We argued that Best Effort

CRediT authorship contribution statement

Attila Vidács: Software, Validation, Visualization, Investigation, Writing - review & editing. Géza Szabó: Conceptualization, Methodology, Visualization, Writing - original draft.

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.

References (30)

  • Robot operating system

    (2019)
  • Gazebo robot simulator

    (2020)
  • ARIAC Rules

    (2020)
  • SucanI.A. et al.

    Moveit!

    (2020)
  • DornhegeC.

    Task Planning for High-Level Robot Control

    (2015)
  • Cited by (5)

    • Design and implementation of agent-based robotic system for agile manufacturing: A case study of ARIAC 2021

      2022, Robotics and Computer-Integrated Manufacturing
      Citation Excerpt :

      The ARIAC requires the robotic system to handle the dynamic order requirement and dynamic environment, which is full of uncertainties (unexpected orders, unknown and arbitrary parts, and unexpected disturbance events). The rule of thumb for ARIAC robotic system design, “prepare mostly for errors instead of normal operation” [9], shows the difference from normal industrial application. Off-the-shelf conventional planning strategies are not suitable due to uncertainty challenges.

    • Context-aware scheduling and control architecture for cyber-physical production systems

      2022, Journal of Manufacturing Systems
      Citation Excerpt :

      In other words, these systems cannot perform their actions in unpredicted situations because they can react only for deterministic and predefined situations without activeness that can perceive the contextual situation and take actions to cope with uncertain disturbances. To advance the agility of industrial robotics assembly systems in unstructured and dynamic environments, the National Institute of Standards and Technology (NIST) collaborated with the Open Source Robotics Foundation (OSRF) to launch the Agile Robotics for Industrial Automation Competition, ARIAC, in 2017 [7]. ARIAC allows global competitors to take the latest technologies and approaches to address practical industrial challenges such as that (1) robots can be replaced without extended downtime or reprogramming, (2) replanning can occur quickly when a new order is received, or (3) responses can be made to dynamic environmental conditions [8].

    • A Container-Based Framework for Developing ROS Applications

      2022, IEEE International Conference on Industrial Informatics (INDIN)
    View full text