Shanghai-based robotics firm Fourier Intelligence on Monday released an open-source dataset that it hopes will allow companies and research institutions around the world to better train robots in the human-like usage of their hands.
Dubbed Fourier ActionNet, the landmark dataset was compiled by having real humans wear virtual reality headsets as they guide a humanoid robot through the use of its hands through a teleoperation system. According to Fourier, this method aims to allow robots to move away from traditional gripper-based systems and instead take advantage of the dexterity of the human hand.
The initial dataset comprises more than 30,000 high-quality training entries, featuring intricate hand movements and specialized imitation learning data for tasks involving hand dexterity. These entries span a wide range of real-world applications, including activities such as picking up and setting down tools, performing household chores, and executing various other hand-related tasks.
The release aims to foster innovation and collaboration among the global robotics community and enhance AI robot training, the company announced on Monday.
Up to now, Fourier has collaborated with over 20 top domestic and international research institutions and leading industry enterprises. The company has pledged it will continue to release more advanced data modules covering full-body motion control and multi-task coordination.
Shanghai-based robotics firm releases open-source dataset on humanoid robot hand usage
