AI Robot System Developed in Collaboration with JAXA Achieves 100% Accuracy in Zipper Operations, Including Curves on Flexible Objects
– Efforts to Automate AI Robot Operations at Manned Space Facilities for Astronaut Work Efficiency-
ExaWizards Inc. (Headquarters: Minato-ku, Tokyo; Representative Director & President: Makoto Haruta; hereafter, “ExaWizards”) is collaborating with the Japan Aerospace Exploration Agency (Headquarters: Chofu-shi, Tokyo; President: Hiroshi Yamakawa; hereafter, “JAXA”) to develop a versatile imitation learning platform to train in various tasks.
This platform is a part of our effort to automate crew operations at manned space facilities. We are pleased to announce that we have used this platform to construct an AI robot system to perform tasks such as opening and closing zippers on cargo bags at fixed locations with 100% accuracy. This task is difficult for autonomous robots to perform using pre-existing technologies.
(Image provided by: JAXA/NASA)
☑︎︎The Challenge of Flexible Objects
External robotic arms contribute to the International Space Station (ISS) by reducing extra-vehicular activity risks and increasing astronaut work efficiency. We must increase their ability to manipulate flexible materials in order to assist future astronauts.
Traditional robotic technologies have difficulty manipulating objects with changing shapes, such as Cargo Transfer Bags (CTB) and cables, in environments with communication delays. The size and weight of these objects also affect the accuracy of these operations.
To address these challenges, we have developed a versatile imitation learning platform that can be trained in various tasks by our “exaBase Robotics” robot AI solution, which uses multimodal data to automate and optimize these processes. We conducted an applicability evaluation and a demonstration on a real machine supported by this platform’s AI robot system to open and close CTB zippers.
☑︎Achieving 100% Accuracy in Zipper Operations on Curved, Flexible Objects
Operating zippers on flexible objects is a task that requires advanced imitation learning and predictive learning technologies, which has been difficult to achieve with traditional technologies. A paper by Professor Tetsuya Ogata, a faculty member of the Department of Applied Engineering at Waseda University’s School of Fundamental Science and Engineering and advisor to our company, demonstrated that it is possible to open zippers on a straight line with 93% accuracy on fixed objects by combining visual and tactile information.
We developed this initiative based on Professor Ogata’s model architecture. It achieved a more practical level of AI robotic zipper operations on flexible objects in the following three aspects:
(1) Achieved zipper operations on flexible objects via force sense (torque value estimated from joint motor current) instead of touch.
(2) Trained the robot to operate straight and curved zippers on flexible objects at fixed locations with a 100% accuracy rate.
(3) Trained the robot by shifting the position of the fixed object and still achieved an over 80% accuracy rate in operating zippers.
We intend to develop this technology to achieve higher accuracy under the changing conditions of space, such as by changing the location and shape of the object and the gravity of the environment.
☑︎AI Robot System Implemented on ROS Base, Capable of Adapting to Future Robot Changes
Systems that depend on specific hardware or software constrain future system updates and increase development time and costs. Our AI robot system is based on the extendable and versatile open-source robot software platform “ROS (Robot Operating System)” and the open platform “Docker,” which can adapt to changes in the development environment. This allows for consistent system operations and keeps time and development costs down for future system updates.
ExaWizards will utilize their “exaBase Robotics” technology and expertise to develop a sustainable space industry by automating tasks in manned space activities. We will also continue to address social issues, including improving the productivity of Japanese companies, by applying the AI and robotics technology and expertise gained through this initiative.
☑︎Comments
Professor Tetsuya Ogata, Department of Applied Engineering, School of Fundamental Science and Engineering, Waseda University
Our “Deep Predictive Learning” imitates the complex manipulations of viscous fluids and flexible objects by combining vision, touch, and force senses. This joint research between ExaWizards and JAXA is an interesting application that maximizes these characteristics. I have great expectations for future developments.
☑︎What is “exaBase Robotics”?
”exaBase Robotics” is an AI solution for robots that automates and optimizes operations utilizing multimodal data such as image data, on-site equipment/robot control data, and simulator creation data. It can be used in many different situations, e.g., when you need an alternative for a simple task, or when you want to replicate the skill of an expert.
exaBase Robotics site: https://exawizards.com/exabase/robotics/
[ExaWizards Corporate Profile]
Company name: ExaWizards Inc.
Location: 21F, Shiodome Sumitomo Bldg., 1-9-2, Higashi Shimbashi, Minato-ku, Tokyo
Representative: Ko Ishiyama, Representative Director & President
Description of business: Industrial innovation and the resolution of social issues through the development of AI-enabled services
URL: https://exawizards.com/
<Contact for public relations>
E-mail address of the Public Relations Division of ExaWizards Inc.: publicrelations@exwzd.com
*ExaWizards is looking for business development/BizDev members who are interested in bringing about change in the world through robots and related multimodal technologies. If you are interested, please contact us via this link<https://open.talentio.com/r/1/c/exwzd/pages/81360>