IEEE Robotics & Automation Magazine - June 2023 - 98

second mission was to use the parts of IKEA's Stefan chair but
assemble the Stefan chair in a deformed shape rather than the
original shape. The instructions for assembling the deformed
Stefan chair were provided at the start of the second mission.
The participating teams were required to use an AI-based
algorithm to understand the instructions for assembling the
deformed chair and generate the correct assembly sequence
for the robot system to execute.
Our team (Team SK2Y) was awarded first place in this
competition because of our effective use of an AI-based algorithm
and a novel robot system for the assembly process. The
Faster-RCNN and the MORAN were applied to recognize the
assembly instructions, and a connection relationship tree algorithm
was developed to interpret the recognized commands.
This AI-based algorithm quickly generated commands that
the robot system could execute. In addition, the implementation
of multifingered grippers and a manipulation station
provided additional advantages during the assembly process
regarding assembly time and convenience. Owing to the use
of these two unique technologies, the assembly process was
successfully executed with high precision and safety, and we
were able to outperform the other teams because of our exceptional
execution time. An overview of the working process is
presented in Figure 1.
FURNITURE ASSEMBLY AI-ROBOT CHALLENGE 2021
COMPETITION SCHEDULE AND RULES
The " Furniture Assembly AI-Robot Challenge 2021 " evaluated
the combined AI's recognition and interpretation ability
with robot automation assembly technology. Four teams participated
in the competition, and each team developed its own
system during the two-year preparation period. The event was
originally scheduled for 2020, but it was moved to March
2021 because of the COVID-19 pandemic. This competition
was held at COEX in South Korea, and after a three-day
installation and operation test period, it was held offline on
the fourth day and broadcast live online. Figure 2(a) shows a
screen capture from the live broadcast on competition day.
The competition ran all day, with the first mission in the
morning and the second in the afternoon.
The principal regulations related to the progress of the competition
were as follows. The worktable on which the chair
was to be assembled needed to be configured within an area
of 2.5 # 2.5 m. Two or more vertical articulated robots were
to be used for assembly, and the type of robot could be freely
selected. In addition, the camera required for recognizing the
assembly process could be installed freely. Before assembly,
the parts to be assembled were voluntarily placed by the team
and inspected by the judges. A team that wanted to gain additional
kitting points applied a kitting action. The team that
applied for the kitting action was given a 400 # 400-mm connecting
part plate with connecting parts randomly placed by
the judge. Before the mission began, a bitmap file with the
assembly instructions was provided via USB, and immediately
after that, the assembly instructions had to be recognized and
assembled within 90 min.
The evaluation index of the competition consisted of
86 points, including parts-fastness achievement (46 points),
completeness (20 points), and speed (20 points). One judge was
Stefan
RecognizeInterpret
Input Assembly Instruction
Recognition With Faster-RCNN and MORAN
Execute
Apply
PDDL Command
Connection Relationship Tree
Real-World Assembly
Robot
AI
FIGURE 1. The whole working process. We developed an AI algorithm that can recognize the essential information in the assembly
instructions, create a connection relationship tree, and then convert it into a PDDL command. The robot assembly system received the
PDDL command and executed the assembly task in the real world. PDDL: planning domain definition language.
98 IEEE ROBOTICS & AUTOMATION MAGAZINE JUNE 2023

IEEE Robotics & Automation Magazine - June 2023

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2023