Chengdu innovation team achieves national breakthrough in giving humanoid robots a "stronger brain"
2024-08-12
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
On August 12, the reporter learned from the Chengdu Humanoid Robot Innovation Center that the center recentlyReleased the latest research and development results: China's first robot multimodal model RRMM (Raydiculous Robot Multimodal Model) and two-arm collaboration system RTACS (Raydiculous Two-Arm Cooperation System). It is reported that the multimodal model (RRMM) can enable the robot to understand and reason about abstract semantic instructions, and dispatch the dual-arm collaborative system (RTACS) to perform tasks. Currently, only foreign leading companies such as OpenAI have released similar results, which can enable humanoid robots to bid farewell to simple "remote control" and have a "stronger brain" and autonomous execution capabilities.
Multimodal (multiple heterogeneous modal) data collaborative reasoning can integrate two or more different perception sources, perform comprehensive reasoning, and complete complex tasks. According to the relevant person in charge of the Chengdu Humanoid Robot Innovation Center, for example, when the latest ChatGPT4o demonstrates its powerful functions, the tester does not use the traditional keyboard to input question instructions, but instead writes an equation to be solved on a piece of paper, draws a geometric figure by hand, and uses voice instructions to ask GPT4o to solve the problem. GPT4o uses image recognition, voice recognition, semantic understanding and reasoning to quickly give accurate answers. This is a typical multimodal application. "The application of multimodality in the field of humanoid robots enables humanoid robots to integrate multiple factors such as images, semantics, force perception, and environmental perception, comprehensively judge, generate tasks, and execute tasks. This is the key core technology for humanoid robots to have autonomous thinking capabilities." The relevant person in charge of the innovation center said.
At present, the first Chinese robot multimodal model released by the Chengdu Humanoid Robot Innovation Center can relatively clearly distinguish and sort fruits and non-fruits, beverages and non-beverages, and can relatively abstractly distinguish and sort cylindrical and rectangular objects, and can relatively abstractly distinguish and sort edible and inedible items. According to reports, to complete the above tasks, the robot needs to autonomously extract and summarize all items through the multimodal model, especially for multiple objects with similar bottle shapes, it is necessary to further comprehensively judge whether they are edible through the text, pictures and other factors on the bottle packaging, rather than non-edible bottled objects such as cosmetics and detergents, in order to finally complete the task, showing that the Chengdu Humanoid Robot Innovation Center has made further breakthroughs in technology.
It is understood that on April 12 this year, Chengdu Humanoid Robot Innovation Center Co., Ltd. obtained a business license, marking the official establishment of the first new humanoid robot research and development institution in the central and western regions. The innovation center focuses on the core technology research, achievement transformation and industrialization of the humanoid robot industry, and promotes the high-quality development of the Chinese humanoid robot industry.
Red Star News reporter Wang Junfeng pictures and videos According to Chengdu Humanoid Robot Innovation Center
Edited by Li Yuyi
(Download Red Star News and get a reward for reporting!)
