OMRON Technology and Intellectual Property H.Q. and OMRON SINIC X Presented Four Research Papers in IROS 2023, a Top-Tier Conference on Robotics

• October 23, 2023

The Technology and Intellectual Property H.Q. of OMRON Corporation (HQ: Shimogyo-ku, Kyoto; President and CEO: Junta Tsujinaga; hereinafter "OMRON") and OMRON SINIC X Corporation (HQ: Bunkyo-ku, Tokyo; President and CEO: Masaki Suwa; hereinafter "OSX") are pleased to announce that four of our research papers (one from OMRON Technology and Intellectual Property H.Q. and three from OSX) have been accepted for publication at the 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (hereinafter "IROS 2023"), a leading international conference in the field. The papers were presented at IROS 2023, which was held in Detroit (US), on October 1-5, 2023.

IROS is one of the world’s largest international conferences on robotics. At IROS 2023, 43.3% of 27,760 paper submissions were accepted.

The Technology and Intellectual Property H.Q. and OSX are engaged in research towards a future where machines harmonize with people to help bring out their creativity and potential. For IROS 2023, we presented research findings on robotic technologies, including a fast computing algorithm for safe robot operation, a soft robot with the ability to autonomously learn assembly tasks, and a robotic system for automating chemical and materials science experiments.

Some of these research findings have been released in articles with a simple explanation of the corresponding technology and on GitHub as open-source code so that they can be actively used for social implementation. For details, please access the link for each paper.

The Technology and Intellectual Property H.Q. will accelerate the resolution of social issues through our core technology of "Sensing & Control + Think" to realize OMRON’s aim of "Empowering People Through Automation." OSX continues to develop value creation via technological innovation through collaboration with universities and external research institutions as well as the recruitment of interns.

<Publication by the Technology and Intellectual Property H.Q.>
*All presentation times are in local time.
ITIRRT: A Decoupled Framework for the Integration of Machine Learning into Path Planning

Authors	Thibault Barbie (OMRON), Shigeharu Mukai (OMRON)
Time of presentation	14:36-14:42 / 15:30-17:00, Wednesday, October 4, 2023
Overview	Effective path planning is essential to ensuring the safe operation of articulated robots by avoiding collisions with people and objects in their surroundings. However, it may not be easy to realize complicated motions in the real world, as computing path planning takes a great deal of time. To address this challenge, a machine learning-based algorithm has been suggested, but it has been difficult to mathematically guarantee success. This study proposes a framework that decouples path planning into the prediction phase by machine learning and the path search phase. This enables the predicted path to be used as an initial proposal, guaranteeing successful path planning while dramatically reducing computation time. Furthermore, the framework’s simple structure makes integrating various machine-learning models easy.
Details	https://github.com/thibaultbarbie/itirrt

This paper contains findings from research in the robotics technology field conducted as part of the technology trial program, which is a Technology and Intellectual Property H.Q. unique system that supports prior research/verification of technology in an effort to create new value.

<Publication by OSX>
*All presentation times are in local time.
Learning Robotic Powder Weighing from Simulation for Laboratory Automation

Authors	Yuki Kadokawa (Nara Institute of Science and Technology; OSX intern from April 2022 to September 2022), Masashi Hamaya (OSX), Kazutoshi Tanaka (OSX)
Time of presentation	14:18-14:24 / 15:30-17:00, Monday, October 2, 2023
Overview	This study aims to achieve milligram-level powder weighing by robotic spoon manipulation for laboratory automation of chemical experiments. The focus is on powder weighing with machine learning to adapt to the fluid dynamics of the powders and the huge variations in fluidity between materials. However, the data collection is impractical in the real world because of powder scattering during the learning process and the effort required to clean up the working environment. This study involved the application of sim-to-real transfer learning with Domain Randomization and the formulation of powder weighing with the spoon manipulation as a reinforcement learning task. The experimental results confirmed that the robot could accurately perform powder weighing even for different materials without additional training in the real world. This led to the realization of the robotic power weighing system, which observes changes in powder weight, without any data collection in the real world.
Details	https://omron-sinicx.github.io/powder-weighing/ https://medium.com/sinicx/learning-powder-weighing-manipulation-4ce2f02fd3e

Learning Robotic Assembly by Leveraging Physical Softness and Tactile Sensing

Authors	Joaquín Royo-Miquel (OSX; OSX intern from January 2022 to March 2023), Masashi Hamaya (OSX), Cristian Camilo Beltran-Hernandez (OSX), Kazutoshi Tanaka (OSX)
Time of presentation	14:36-14:42 / 15:30-17:00, Tuesday, October 3, 2023
Overview	The aim of this study is to achieve robotic assembly in uncertain environments where there are variations in the grasping poses of parts and imprecise target positions caused by observation and modeling errors. In these environments, the robots must recognize the surrounding environments, such as the part or target positions, and perform assembly by adapting to these variations. This study proposed a method wherein a robot, equipped with a physically soft wrist and tactile sensors, performs assembly while exploring the environment. The physically soft wrist allows for safe contact and exploration, and the tactile sensor can capture contact events as insertion parts contact the holes. As a result, the robot performed peg-in-hole tasks more robustly by adapting to uncertainties in the target position, variations in peg diameter, and misalignment of the peg grip. This method thus enabled robots to perform effectively in uncertain environments without complex configurations. Furthermore, the method is expected to significantly reduce labor for deploying robots and to contribute to the spread and easy implementation of robots.
Details	https://omron-sinicx.github.io/saguri-bot-page/ https://medium.com/sinicx/learning-soft-robotic-assembly-using-tactile-sensing-20470b22e943

Robotic Powder Grinding with Audio-Visual Feedback for Laboratory Automation in Materials Science

Authors	Yusaku Nakajima (SOKENDAI), Masashi Hamaya (OSX), Kazutoshi Tanaka (OSX), Takafumi Hawai (Osaka University), Felix Wolf Hans Erich von Drigalski (Mujin Inc.), Yasuo Takeichi (Osaka University), Yoshitaka Ushiku (OSX), Kanta Ono (Osaka University)
Time of presentation	09:18-09:24 / 10:00-11:30, Wednesday, October 4, 2023
Overview	Powder grinding is essential for sample preparation in materials science experiments. However, there was a challenge that it required a great deal of manpower and time in grinding a few grams of powder to the desired size for the experiments. To address this challenge, we have previously developed a robotic powder grinding system that can automatically decide whether to gather or grinding process by recognizing the distribution of powder in a mortar based on visual feedback, thereby improving the efficiency of powder grinding. Nevertheless, it is difficult to estimate the powder particle size during the grinding process using only visual feedback, and there have been cases where the powder could not be ground to the desired size. This study proposed a multi-modal robotic powder grinding system that incorporated visual feedback as well as audio feedback. This system can estimate the particle size by audio intensity, which makes it possible to perform appropriate grinding by evaluating grinding progress. Experimental results demonstrated that the system helped to grind much smaller the particle sizes than in a previous study by using both visual and audio feedback. The system will make a significant contribution to the preparation of a wide variety of samples in small quantities and the analysis of powder grinding processes in materials science experiments.
Details	https://omron-sinicx.github.io/powder-grinding/

About the Technology and Intellectual Property H.Q.
The Technology and Intellectual Property H.Q., OMRON Group’s corporate R&D division, is working to anticipate social issues that may arise in the near future and create innovation driven by social needs to solve them, while evolving our core technology of "Sensing & Control + Think." With the aim of expanding human potential and realizing a future in which people can play a more active role, we are conducting a variety of research and development activities in areas such as AI, robotics, energy management, and sensing based on our founder Kazuma Tateishi’s management philosophy: "To the machine, the work of the machine, to man the thrill of further creation." For more details, please refer to https://www.omron.com/global/en/technology/

About OMRON SINIC X Corporation
OMRON SINIC X Corporation is a strategic subsidiary seeking to realize the "near future designs" that OMRON forecasts. Researchers with cutting-edge knowledge and experience across many technological domains, including AI, Robotics, IoT, and sensing, are affiliated with OSX, and with the aim of solving social issues, they are working to create near future designs by integrating innovative technologies with business models and strategies in technology and IP. The company will also accelerate the creation of near future designs through joint research with universities and external research institutions. For more details, please refer to https://www.omron.com/sinicx/en/