机器人大讲堂·2025-05-23 12:11Group 1 - Tesla released a new video showcasing the humanoid robot "Optimus" performing various tasks such as throwing garbage, cooking, vacuuming, and ironing, all accomplished through a single neural network that learns directly from human videos [1] - Elon Musk reiterated that the Optimus humanoid robot will become one of Tesla's most important products during a recent interview [1] Group 2 - Apple introduced a new training method called "PH2D" that combines human coaches and robot demonstrators to enhance the efficiency and cost-effectiveness of humanoid robot training, addressing the limitations of traditional methods [2][4] - The new approach utilizes modified consumer-grade devices to create training materials, and Apple developed a model named "Human-humanoid Action Transformer" (HAT) to process data from both human coaches and robot demonstrators [4] Group 3 - A collaboration between Harbin Institute of Technology and Dongfang Electric Group achieved significant advancements in soft robotics, developing a smart soft robot based on MXene materials for special environments [5][7] - The research emphasizes the integration of material innovation and structural design, providing new technical means for the development of three-dimensional soft robots [7] Group 4 - French retail giant Leclerc's president called for robots and AI industries to contribute to social security funding, suggesting a "robot tax" to address the financial strain on the social security system caused by technological advancements [8][10] - The proposal aims to create new funding sources and improve employee wage levels in response to the increasing pressure on social security funds [10] Group 5 - Researchers from Sungkyunkwan University reported a new type of artificial synaptic mechanical receptor array that significantly reduces data burden while improving recognition accuracy in tactile sensing applications [11][13] - The study demonstrated that the new method achieved a peak accuracy of 92.3% in texture recognition tasks using only 10.6% of the data [13]