Transforming Humanoid Learning Shapes
Industrial physical artificial intelligence (AI) startup Diden Robotics and human motion AI startup Mubin announced on the 15th that they have signed a business agreement to build and utilize datasets necessary for humanoid learning.
Through this agreement, the two companies plan to increase the robot's motion learning performance together. It combines Mubin's LiDAR-based motion capture technology with multimodal data collection capabilities with Didenrobotics' humanoid development capabilities.
The two companies plan to build a data pipeline that collects various human movements in real-world environments with high-precision 3D motion data and multimodal sensor data and converts them into a form suitable for humanoid robot learning.
Existing motion data are often built for the purpose of content creation or animation, so it is difficult to apply them as they are due to the joint structure, motion range, and physical constraints of the robot.
Mubin collects lidar and camera data from various perspectives and provides it in a retargeted form to suit the robot structure of Diden Robotics. Using this dataset, DidenRobotics will increase the learning efficiency and precision of key behavior models, including humanoid walking, balancing, systemic control, and interaction with objects and the environment.
Starting with this MOU, the two companies plan to expand the scope of cooperation to the possibility of joint research and commercialization.
Kim Jun-ha, CEO of Diden Robotics, said, "In order to improve humanoid performance, we need high-quality multimodal data containing human movement and spatial information," adding, "We will secure data optimized for robot hardware and strengthen a learning system that can be applied directly to actual robot development."
"The technology that precisely collects human movements and converts them to fit the robot's body structure is an important foundation for humanoid robot learning," said Choi Byul-yi, CEO of Mubin. "We look forward to expanding Mubin's motion data technology beyond the content industry to the core data infrastructure in the robotics field."
