Kazushi Yamamoto, Sehwa Chun, Yuki Sekimori, Chihaya Kawamra
On August 28 and 29, students and enthusiasts gathered online to take part in the Underwater Robot Convention in JAMSTEC 2021. The event, hosted by NPO Japan Underwater Robot Network, serves as a forum for participants to exchange technical ideas and build networks through presentations and the competition of underwater robots. The overview of this year’s event can be found in the convention’s website  (in Japanese), and the events in previous years are described in , , , and .
We, the authors, are masters course students in Prof. Maki’s laboratory at the University of Tokyo. We participated in the AI Challenge division as team “UT Maki Lab.” [Figure 1]. Our primary aim was to accumulate basic knowledge and skills related to underwater robotics in preparation for future research work. We also learned how to collaborate as a team, which is a crucial aspect in developing and operating underwater robots.
In the AI Challenge division, participants proposed a mission that combined underwater robots and Artificial Intelligence (AI). The event organizers called for “exciting” robots that are “capable of executing AI missions underwater.” Each team uploaded a poster and two videos of five minutes each, one to present their robot and AI mission, and another to demonstrate its capability. The judges evaluated the overall performance based on the four criteria listed in table 1.
Table 1. Criteria for AI Challenge division
|Presentation||20||The overall quality of presentation, videos and poster.|
|Idea||30||The novelty of the AI mission.|
|Technical contents||30||The quality of hardware, software, algorithms, etc.|
|Capability||20||The performance of the robot. Maximum points for full autonomy (without tether cables).|
Given the freedom to choose our own AI mission, we wanted to set an exciting and technically interesting topic. Currently, the world is under the threat of COVID-19, so we thought that a mission related to it would be attractive. Based on our slogan, “No exception to prevent infection,” our mission was to identify the faces without a mask with a computer vision (CV) algorithm and cover the faces with a mask with artificial intelligence using an underwater robot. We installed three portraits and a mask station in the 8m on a side cubic water tank in the Institute of Industrial Science, the University of Tokyo [Figure 2]. One of the portraits wore a mask, and the other two did not. The funnel-shaped mask station aligns the robot, such that it can dock regardless of a small position error.
Our AI mission was to cover all the faces not wearing a mask. The robot searches and detects a maskless face within the set of submerged photograph portraits and delivers a mask from the docking station. This mission was divided into tasks: detect, catch and delivery. During the detect task, the robot hovers and searches for the portraits using the CV algorithm with the AI. If the probability of the portrait wearing the mask is low, the robot decides to transition to the next task. During the catch task, the robot aims to dock on the mask station and fetch the metallic mask with the electromagnet. During the delivery task, it delivers the mask to the portrait identified in the detect task. After the robot puts the mask on the target, it transitions back to the detect task to search for the remaining maskless portrait. If everyone is wearing a mask, the robot completes the mission, and it ascends to the surface.
To demonstrate our mission, we developed a hovering type autonomous underwater vehicle named “AI and ROS Integrated vEhicLe” (ARIEL) [Figure 3, Figure 4]. ARIEL estimates its depth using a depth sensor that measures water pressure. It estimates its position on the horizontal plane by perceiving the markers attached to the station using a camera [Figure 5]. It equips an electromagnet on the bottom to transport the masks.
ARIEL controls itself using the Raspberry Pi microcomputer and Teensy drivers. We implemented the software applications of ARIEL using the ROS, an open-source library and tools commonly used to develop robotics applications. The ROS is a powerful tool for parallel distributed processes such as sensor and actuator signal handling. The convolutional neural network of ARIEL can estimate the position of the faces on the bottom of the water tank. Furthermore, it can calculate the probability of whether the person in the portrait is wearing a mask or not. We created the AI software based on the Tensorflow library, and we used the MobilenetV2 neural network  for mask recognition [Figure 6]. This software program runs swiftly and accurately on the Raspberry Pi despite its limited computational power. ARIEL decides its action based on its state and the position of the photograph. It completes the mission accurately by transitioning to an appropriate mode depending on the situation.
It was June, 2021, when we started working in earnest. With less than three months left until the date of submission on 20 August , we needed to progress efficiently. We divided the ARIEL’s development process into three subprojects: hardware, system software, and AI algorithm, and we divided and conquered the subprojects. Each process was carried out in parallel by appointing a person in charge of each part. While accomplishing each part of the task, it was important to collaborate and exchange opinions as well. In July, the hardware was completed to some extent, and it was possible to operate ARIEL with manual control. In the beginning of August, we combined the system and AI that were in progress at the same time with the hardware, and the prototype of ARIEL was completed.
From August, we tested ARIEL in the water tank and made adjustments. We adjusted the center of buoyancy and the center of gravity, tuned the PID parameters, and set the parameters of the AI detector. Before the submission date, ARIEL was completed as an autonomous underwater robot and capable of judging whether a mask is on and transporting the mask by itself [Figure 7, Figure 8]. ARIEL showed 53% mission success rate by succeeding 9 times out of 17 trials.Result
On the first day of the event, each team shared a presentation video and answered questions online. On the second day, each team shared a demonstration video likewise, and we had the awarding ceremony in the end. Our videos can be found in our laboratory YouTube channel , and our final score is shown in Table 2. With an outstanding score of 91 points, we came in first place out of three teams.
Looking back, the Underwater Robotics Convention made us realize the difficulty of developing an underwater robot, the importance of collaboration, and connected us with people in the related industry. We have no doubt that the experiences gained throughout the intense three months of development, and the event, will be of great strength for us in future research work.
Table 2. Final scores for “Team UT Maki Lab.”
After the event
Shortly after the Underwater Robotics Convention 2021 in JAMSTEC, “UT Maki Lab.” and a team from Fukushima Prefectural Taira Technical High School were invited to demonstrate our robots in Roboichi , an event that took place in Fukushima Robot Test field along with the World Robot Summit from October 8-10. ARIEL operated for over 10 hours under manual control, and did not experience any technical issues. It demonstrated the ability to detect maskless, and intervene with detected targets by placing the metallic mask on and off the portrait.
Finally, we will close this article with a short comment from each member.
Kazushi Yamamoto: Making ARIEL with my mates was great fun. It was an honor (and a relief) to come in first place.
Sehwa Chun: It’s really cool to work with people with high enthusiasm, diverse talents and the same goals. It couldn’t be greater if it was about underwater robots!
Yuki Sekimori: I would like to appreciate all my teammates for the outstanding enthusiasm, collaboration, and performance throughout the project. I would also like to appreciate the members of the Maki Laboratory for the pieces of advice and encouragement. Finally, I would like to thank all participants, event organizers and sponsors for providing us with the wonderful experience.
Chihaya Kawamura: This project was not only challenging but also exciting for me because it was my first time facing robot development everyday. The most striking thing I learned through this project is that haste makes waste. We found out that when ARIEL goes wrong for unknown reasons, the fastest way to fix it is to go through all the possible reasons one by one.
The Underwater Robot Convention in JAMSTEC in 2021 was supported by The Japan Society of Naval Architects and Ocean Engineers, IEEE/OES Japan Chapter, MTS Japan Section, Techno-Ocean Network, Kanagawa Prefecture, Yokosuka City, Tokyo University of Marine Sciences and Technologies, Japan Agency for Marine-Earth Science and Technology (JAMSTEC), Center for Integrated Underwater Observation Technology at Institute of Industrial Science, the University of Tokyo, The Nippon Foundation, Misago Co., Ltd., FullDepth Co., Ltd., Aqua Modelers Meeting, and Matsuyama Industry Co., Ltd. We would like to express our sincere appreciation to the sponsors for their strong support and cooperation in realizing this event.
 Underwater Robot Convention in JAMSTEC 2021 (Japanese). http://jam21.underwaterrobonet.org/
 Y. Sekimori, T. Maki, Underwater Robot Convention in JAMSTEC 2020 – All Hands on Deck! Online!!, IEEE OES Beacon Newsletter, 10(1), 39-42 (2021.3)
 K. Fujita, Y. Hamamatsu, H. Yatagai, Reflection for Singapore Autonomous Underwater Vehicle Challenge – the Comparison Between SAUVC and a Competition Held in Japan, IEEE OES Beacon Newsletter, 8(2), 64-67 (2019.6)
 H. Yamagata, T. Maki, Underwater Robot Convention in JAMSTEC 2018 – from an Educational Perspective, IEEE OES Beacon Newsletter, 7(4), 68-72 (2018.12)
 H. Horimoto, T. Nishimura, T. Matsuda, AUV “Minty Roll” and results of “Underwater Robot Convention 2017 in JAMSTEC,” IEEE OES Beacon Newsletter, 6(4), 77-79 (2017.12)
 M. Sandler, A. Howard, M. Zhu, A. Zhmoginov and L. Chen, ”MobileNetV2: Inverted Residuals and Linear Bottlenecks,” 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, pp.4510-4520, DOI: 10.1109/CVPR.2018.00474.
 Maki Laboratory Official YouTube channel. https://www.youtube.com/user/makilabo
 Roboichi (Japanese). https://biz.nikkan.co.jp/roboichi/