Core Viewpoint - Xinghaitu (Beijing) Artificial Intelligence Technology Co., Ltd. has completed its stock reform and is the first company in the embodied intelligence sector to do so this year [1] Group 1: Company Overview - Xinghaitu was established in September 2023 and has completed seven rounds of financing since its inception, with investments from major institutions such as Ant Group, Meituan, Baidu Ventures, Lenovo Capital, Hillhouse Capital, and IDG Capital, achieving a valuation of over 10 billion yuan [2] - The company focuses on the "one brain, multiple forms" embodied intelligent robot field, achieving a full-stack self-research from embodied entities, end-to-end AI algorithms to scene solutions, and has built a complete business layout of "entity + platform + model + data" [4] Group 2: Product and Market Development - Xinghaitu's product lineup includes core products such as the general humanoid robot R1 Pro and the wheeled dual-arm mobile platform R1 Lite [4] - The company has launched the G0 Plus VLA model, which has seen over 500,000 downloads of its open data set, indicating strong interest and engagement [4] - In terms of commercialization, Xinghaitu has secured large-scale orders in the thousands from leading domestic automotive manufacturers and logistics companies, with business scenarios covering key sectors such as manufacturing and logistics [4]

点击下方 卡片 ，关注" 具身智能 之心 "公众号 >> 点击进入→ 具身智能之心 技术交流群 更多干货，欢迎加入国内首个具身智能全栈学习社区 ： 具身智能之心知识星球 (戳我) ， 这里包含所有你想要的。 Omdia近段时间发布的2025年全球人形机器人出货量报告，引起了一场不小的风波。多家媒体转发，其中排行榜单上显示智元机器人以5168台作为榜首，中 国企业出货全球占比超过80%。从数量上来看，出货量在500套及以上的公司一共有四家：智元机器人、宇树、优必选，乐聚机器人（都是中国企业）。 虽然国内企业遥遥领先海外，但作为人形机器人领域的巨头宇树科技对数量有疑议，昨日在官方微信公众号上正式发文澄清："并未对外透露任何销售数 据，我们的人形机器人交付早已超5500台，订单则更多！"。 关于宇树2025年销量数据的澄清 2026年1月22日 | Table 1: Robotics shipments by vendor, 2024 and 2025 | | | | --- | --- | --- | | Vendor | 2024 | 2025 | | AGIBOT | 600 | 5,168 | | Unit ...

Core Viewpoint - The article emphasizes the importance of continuous evolution and adaptability in robotics, particularly through the introduction of the Imeta-Y1, a lightweight and cost-effective robotic arm designed for beginners and researchers in the field of embodied intelligence [2]. Group 1: Product Introduction - Imeta-Y1 is designed specifically for novices and researchers, providing a low-cost and efficient solution for algorithm validation and project development [2]. - The robotic arm features high-precision motion control, low power consumption, and an open hardware and software architecture, facilitating seamless integration from simulation to real-world applications [5]. Group 2: User-Friendly Features - The product offers a comprehensive open-source toolchain and code examples, enabling users to complete the entire process from data collection to model deployment [3][17]. - It supports dual programming languages (Python and C++) and is compatible with ROS1 and ROS2, allowing users to quickly adapt regardless of their programming background [3][18]. Group 3: Technical Specifications - The Imeta-Y1 has a weight of 4.2 kg, a rated load of 3 kg, and 6 degrees of freedom, with a working radius of 612.5 mm and a repeat positioning accuracy of ±0.1 mm [8][19]. - The arm operates at a supply voltage of 24V and utilizes CAN communication, with a control method that includes trajectory tracking, teaching, and API [19]. Group 4: Development and Support - The product provides a full-process toolchain for data collection, model training, and inference deployment, supporting multi-modal data fusion and compatibility with major frameworks like TensorFlow and PyTorch [36]. - The company ensures rapid customer support with a 24-hour response time and offers bulk purchase discounts, as well as project development and training services [19][48].

编辑丨 量子位 点击下方 卡片 ，关注" 具身智能之心 "公众号 >> 点击进入→ 具身 智能之心 技术交流群 更多干货，欢迎加入国内首个具身智能全栈学习社区： 具身智能之心知识星球（戳我） ，这里包含所有你想要的! 在具身智能 （Embodied AI） 的快速发展中， 样本效率 已成为制约智能体从实验室环境走向复杂开放世界的瓶颈问题。 不同于纯数字域的对话任务， 具身任务 通常涉及极度复杂的物理环境感知以及高维度的连续控制输出，这意味着智能体面临着巨大的状态- 动作搜索空间，导致学习效率低下且难以收敛。 传统的无模型强化学习由于缺乏对底层物理逻辑的理解，完全依赖于海量的盲目试错来获取学习信号。 然而，在现实物理世界中，每一次交互都伴随着不可忽视的时间损耗、高昂的硬件维护成本以及潜在的安全风险，这使得动辄数亿次的交互 需求变得极不现实。 为了应对这一挑战， 世界模型强化学习 （World Model RL） 研究应运而生。 其核心范式在于通过额外学习一个能够表征环境内在转移规律的预测模型，使智能体具备在想象空间中进行自我进化的能力。 这种机制允许智能体在潜空间内进行大规模、低成本的轨迹预演与策略优化，从而显 ...

Core Viewpoint - The article introduces RLinf as the world's first RL infrastructure designed for training embodied large models, emphasizing its focus on VLA+RL tasks in the embodied domain and future feature expansions [1]. Group 1: Project Overview - RLinf is positioned as a pioneering platform for embodied large model training, with a unique emphasis on VLA+RL tasks [1]. - The project website is available for further information and resources [2]. Group 2: Technical Features - RLinf supports various simulators and real-world robotics applications, including ManiSkill, Franka Arm, and RoboTwin, among others [4]. - The platform incorporates multiple models such as VLA, OpenVLA, and GROOT, alongside a range of algorithms including PPO, SAC, and LoRA SFT [4]. Group 3: Recruitment Opportunities - The project is actively recruiting for master's, doctoral, postdoctoral, and research assistant positions, with opportunities linked to prominent institutions like Tsinghua University [5]. - There are pathways for non-academic participants to be recommended for leadership roles in various companies, offering competitive salaries and cutting-edge research directions [6]. Group 4: Contact Information - Interested parties can reach out to the RLinf project lead, Yu Chao, via email for further inquiries [7].

数采方案也从仿真优先慢慢到UMI和更加拟人的方案演变，让数据能够规模化和好用，是各家公司一直探索 的。任务的差异化，也对数据的生产方式有一定的要求。 相比于传统机器人，具身领域的算法则更AI，从VLA、VLN到交互大模型，从强化学习到世界模型。基于 模仿学习和强化学习的方案，正在让模型变得更加泛化。 最近越来越多的同学开始准备具身方向的求职，算法、开发、仿真、强化、市场、产品等。不约而同的都再 问一句话，有没有好的求职指南？这个行业是怎么样的？ 这件事情，真的值得梳理一下。 2022年，当大多数人还没意识到"具身智能"即将爆发时，少数开拓者已经在悄悄地摸索着具身机器人的数 据、算法和推理。虽然没有达到那座山，但算法和硬件的高度一直在不断提升，场景也逐渐清晰。 本体层面的稳定性和实用性在陆续提升，从简单的双足、四足机器人，到更精美的人形和移动操作机器人。 场景一直决定着机器人的形态，各家零部件厂商也如雨后春笋般成长，强大的供应链让落地的成本不断下 降。 一个做算法的同学说，如果能把具身的上下游、开发流程、场景和商业化都过一遍就好了。开发的时候既知 道目的，又知道成本，游刃有余。不能管中窥豹，只见一斑。 一个想要 ...

Core Insights - The article presents a comprehensive review of Online Diffusion Policy Reinforcement Learning (DPRL), highlighting its potential to enhance robotic control through a unified algorithm taxonomy and benchmarking system [2][30]. Group 1: Challenges in Online DPRL - The integration of diffusion strategies with online RL faces three core challenges: incompatibility of training objectives, high computational costs and gradient instability, and insufficient generalization and robustness [4][5]. - The training objective conflict arises from the inherent incompatibility between the denoising training objectives of diffusion models and the policy optimization mechanisms of online RL [5]. - The computational and gradient issues stem from the multi-step backpropagation required for diffusion models, leading to high computational costs and potential gradient vanishing or explosion [5]. Group 2: Algorithm Classification Framework - The paper proposes a classification framework for Online DPRL algorithms, categorizing them into four main families based on their policy improvement mechanisms [7]. - Action-Gradient methods optimize policies directly through action gradients, avoiding the complexities of diffusion chain backpropagation, with algorithms like DIPO and DDiffPG [9]. - Q-Weighting methods modulate diffusion loss using Q-value weights to guide policies towards high-reward areas, represented by algorithms such as QVPO and DPMD [10]. - Proximity-Based methods approximate the calculation of policy probability densities, enhancing performance in large-scale parallel environments, exemplified by algorithms like GenPO [11]. - BPTT-Based methods utilize end-to-end backpropagation through the entire diffusion process, with algorithms like DACER, but face scalability issues as diffusion steps increase [12]. Group 3: Empirical Analysis and Benchmarking - A unified benchmarking system was established on the NVIDIA Isaac Lab platform, covering 12 robotic tasks to systematically evaluate algorithm performance across five key dimensions [13][15]. - The analysis revealed that GenPO ranked first in 6 out of 12 tasks, while DIPO performed best in offline strategies with an average ranking of 3.58 [15]. - Performance in parallel environments showed that GenPO and PPO significantly improved in larger scales, while DIPO demonstrated robustness across varying parallelization scales [18]. Group 4: Performance and Generalization - The study assessed the impact of diffusion step expansion on performance and latency, finding that Action-Gradient and Q-Weighting methods improved with increased steps, while BPTT methods faced performance declines beyond 20 steps [21]. - Cross-robot generalization tests indicated that offline strategies like DIPO and QVPO exhibited stronger transfer robustness compared to online strategies, which struggled with significant hardware differences [23]. - The robustness of algorithms in out-of-distribution environments was evaluated, with GenPO showing excellent performance in certain scenarios but also a risk of overfitting to source environments [27]. Group 5: Conclusions and Future Directions - The review establishes a theoretical framework for Online DPRL, revealing trade-offs between sample efficiency and scalability, as well as performance and generalization [30]. - Recommendations for algorithm selection include prioritizing GenPO for large-scale simulations, DIPO for resource-constrained scenarios, and Action-Gradient or Q-Weighting methods for high-precision tasks [31]. - Future research directions include integrating safety constraints, exploring multi-agent DPRL, and developing hierarchical RL architectures to enhance exploration efficiency [31].

具身智能之心人形机器人与强化学习技术交流群成立了，欢迎从事RL、人形机器人相关方向的同学加入。 感兴趣的同学添加小助理微信AIDriver005，备注"方向+机构+姓名/昵称"。 ...

Core Viewpoint - The cost of robotic arms has significantly decreased, with prices now below 5000 yuan, making them more accessible for various VLA tasks [1][2]. Group 1: Cost Trends - Two years ago, the price for a single robotic arm for VLA tasks was over 30,000 yuan, which dropped to around 15,000 yuan last year, and now it is below 5,000 yuan [2]. - The reduction in costs allows for easier implementation of various VLA tasks such as pi0 and pi0.5 [2]. Group 2: Challenges for Beginners - Many beginners face difficulties in replicating VLA tasks due to high costs and lack of effective data collection methods [3][4]. - A significant amount of time is wasted by beginners on troubleshooting and overcoming obstacles in data collection and model training [4]. Group 3: Educational Initiatives - The company has developed a comprehensive course aimed at addressing the challenges faced by beginners in the VLA field, covering hardware, data collection, algorithms, and practical experiments [9][14]. - The course includes a free SO-100 robotic arm for participants, enhancing hands-on learning [19]. Group 4: Target Audience and Requirements - The course is designed for individuals seeking practical experience in VLA, including students and professionals transitioning from other fields [26]. - Participants are expected to have a foundational knowledge of Python and Pytorch, as well as experience in debugging and data collection with real machines [26].

现有数据集真的推动机器人能力提升了吗？ 近年来，随着机器人学习和模仿学习的快速发展，各类数据集与方法层出不穷。然而，这些数据集及其任务设计往往缺乏系统性的考量与原则。这引发了两个关键 问题：现有数据集与任务设计是否真正推动了机器人能力的提升？仅通过少数常见任务的评估，能否准确反映不同团队提出的各类方法在不同任务上的差异化性 能？ 为解决这些问题，上海交大等研究团队提出GM-100基准测试，将其作为迈向机器人学习奥林匹克盛会的第一步。GM-100包含100项精心设计的任务，涵盖各类交互 场景与长尾行为，旨在提供一组多样化且具有挑战性的任务集合，全面评估机器人智能体的能力，并推动机器人数据集任务设计向多样化与复杂化方向发展。这些 任务通过对现有任务设计的系统性分析与扩展，并结合人物交互基元与物体功能特性的相关insights开发而成。 点击下方 卡片 ，关注" 具身智能 之心 "公众号 编辑丨具身智能之心 本文只做学术分享，如有侵权，联系删文 >> 点击进入→ 具身智能之心 技术交流群 更多干货，欢迎加入国内首个具身智能全栈学习社区 ： 具身智能之心知识星球 (戳我) ， 这里包含所有你想要的。 本数据集在不同的 ...