According to a CCTV news report, a few days ago, the state local co-constructionembodied intelligence robotInnovation Center and Peking University's School of Computer Science, jointly launched China's first general-purpose body-intelligent robot training open sourceDataset. The dataset collects data on multiple forms of robot ontologies and contains more than two hundred multi-class scenarios with different tasks.
The reporter saw in the robot data collection field of the innovation center that the engineer, by operating the robotic arm, theCapture the robot's completed movement data in a virtual world.. Robots can learn and grasp movements through remote control devices; engineers over here, by wearing full-body motion-capture costumes, can teach robots how to mimic human behaviors and habits one-to-one.
1AI learned from the report that the innovation center has also built out multiple types of robot data training grounds for commercial retail, factories, and home environments. It meets the real environment needs of robots for daily learning and data collection.
According to statistics, the dataset released by the Innovation Center uses a variety of forms of robot ontologies for data collection, including single-arm robots, dual-arm robots, humanoid robots, and other forms of robots.Multi-class scenarios with 279 different tasks, covering up to 61 different objects, with multiple ontologies, skills and applicationsIt is the first body intelligence dataset in China that is generalizable and validated by models.