Group 1 - The core achievement of Harbin Engineering University is the development of an electro-hydraulic deep-sea soft robot, which has been tested in various deep-sea environments, including depths of 1369 meters and 4070 meters [2] - The robot utilizes a miniaturized energy control system for electro-hydraulic unit coordination, enabling it to perform various movements in deep-sea conditions [2] - The soft robot is equipped with a micro deep-sea optical perception system, allowing it to sense its motion state and environmental targets in extreme underwater conditions [2] Group 2 - Figure Company showcased its humanoid robot, Figure 02, which can autonomously fold clothes, utilizing the Helix model for visual language action [5] - The Helix model allows the robot to learn directly from real-world scenarios without relying on rigid fabric models, enhancing its operational capabilities [5] - The robot's neural network has transitioned from merely moving boxes to accurately picking up clothes and folding them through a data-driven approach [5] Group 3 - The 2025 World Robot Conference in Beijing resulted in significant industry promotion, with over 19,000 robots and related products sold, generating sales exceeding 200 million yuan [9] - The conference featured 220 renowned domestic and international robot companies, showcasing 1569 products and securing a total financing amount of 1.481 billion yuan [9] - New initiatives were launched during the conference, including talent cultivation plans and the release of research reports on the development trends of embodied intelligent robots [9] Group 4 - A new data-driven framework called Bioinspired Predictive Slip Control (BPSC) was proposed to enhance robotic manipulation by actively suppressing slip during operations [12] - The BPSC framework integrates neural network predictions with model predictive control, significantly improving stability and adaptability in complex handling tasks [12] - Experimental results indicate that the slip suppression efficiency of this method is 82% higher than traditional gripping force control methods [12] Group 5 - Ninebot Company reported a 76.14% year-on-year increase in revenue, reaching 11.742 billion yuan, and a net profit growth of 108.45%, amounting to 1.242 billion yuan [15] - The company has accumulated 3,790 patents and is actively involved in standardization activities, filling several industry gaps across various technology fields [15] - Key technologies developed by Ninebot include sensorless drive technology and autonomous navigation systems, which have been widely applied across multiple product lines [15]

白交 发自 凹非寺 量子位 | 公众号 QbitAI 注意看，机器人像人一样从容地走出大门了！ 甚至，还有一整支机器人队伍迎面走来。 人形机器人独角兽Figure，再次带来他们的新成果—— 利用强化学习实现自然人形行走 。 跟之前版本的机器人相比，确实更像人了许多，而且步态更加轻盈，速度也更快。 根据官方介绍，主要分成三个部分： 强化学习 ：强化学习利用模拟试验和错误，教Figure 02 人形机器人如何像人一样行走。 模拟训练 ：通过高保真物理模拟器学习如何像人类一样行走，结果只需几个小时就能模拟出多年的数据。 Sim-to-Real ：通过将仿真中的域随机化与机器人上的高频扭矩反馈相结合，模拟训练无需额外调整即可直接转换为真实硬件。 网友们纷纷表示被惊艳到，甚至觉得像是 太空行走 。 有一说一，自从与OpenAI取消合作后，这成果输出确实又快又多。 机器人像人一样自然行走 此次推出的，是经过强化学习训练的端到端神经网络。 具体来看。 首先，利用强化学习技术，在GPU加速物理仿真中对新的行走控制器进行了全面训练，并在几个小时内收集了数年的仿真演示数据。 在模拟器中，数以千计的Figure 02机器人被并行模 ...