Skip to content
#Robot models and learning
Aug 24Mon
  1. OFweek Robotics53

    Swiss startup Flexion demonstrates humanoid robot autonomously completing office package pickup task; commentary says AI models, not hardware, are the key to competition

    According to Wired, Swiss startup Flexion Robotics used a modified Unitree humanoid robot to autonomously open doors, climb stairs, take an elevator, and retrieve and unpack a package in an office. Its approach trains individual skills in simulation, with a master AI model learning task decisions by watching human videos and orchestrating skill combinations, and every layer trained with reinforcement learning. The article cites an ABI Research analyst saying humanoid hardware is becoming commoditized and AI models are the core competitiveness, and contrasts the US bet on models with China's hardware mass-production advantage.

  2. TetraBOT/天创机器人28

    Tianchuang Robotics Showcases T4-A Humanoid Robot and T-WAVES Model at WRC 2026 World Robot Conference

    From August 19 to 23, Tianchuang Robotics appeared at the WRC 2026 World Robot Conference at the China Telecom and CVTE booths, exhibiting the T4-A integrated inspection-and-operation humanoid robot and its self-developed general industrial embodied intelligence fusion model T-WAVES. The T4-A uses China Telecom's RaaS platform to achieve integrated monitoring, recognition, and operation, and can replace personnel in entering 4D scenarios to perform operations and maintenance. At the event, the T4-A equipped with T-WAVES completed multiple live demonstrations of equipment cabinet operations in power scenarios.

Aug 22Sat
  1. Elite Robots/艾利特机器人28

    Elite Robots at WRC 2026: One Show, Three AI+Robotics Deployments

    At the 2026 World Robot Conference (WRC 2026) in Beijing, held August 19–23, Elite Robots presented three AI+robotics application scenarios: Lingju model random bin picking (CS66 collaborative robot + Lingju Zhizao end-to-end VLA model + SMC electric gripper), a dual-arm coffee robot co-exhibited with Heshisiwei (supporting 50+ flavors), and a fully automated ice cream robot co-exhibited with Beijing Rural Commercial Bank.

  2. 银河通用机器人 · X14

    Learning humanoid robot ET1 debuts at WRC

    Not just dancing. ET1 is learning. 🤖✨ Powered by GALBOT's AstraBrain-Agent, ET1 is an intelligent-agent humanoid robot with learning capabilities, able to perceive, understand, think, and act in the physical world. At WRC 2026, ET1 performs live every hour, every day, learning and adapting through interaction. Who has better dance skills? 👀

Aug 21Fri
  1. 众擎机器人 · 微信公众号62

    EngineAI Unveils Human-Brain-Inspired Embodied Intelligence Engine EngineAI Awaken at WRC

    On August 19, 2026, at the World Robot Conference (WRC), EngineAI unveiled EngineAI Awaken, a human-brain-inspired embodied intelligence engine. It is a five-layer architecture comprising the S1 instinctive reflex layer, the S2 motion planning and generation layer, the S3 cognitive reasoning layer, and the S4+S5 self-growth layer (an emotional large model).

  2. Tokyo Robotics · YouTube31

    Dual-arm robot Torobo2 autonomously clears a table in demonstration

    Tokyo Robotics demonstrated its dual-arm robot Torobo2 autonomously clearing a table (bussing): it collects dishes and a cloth from the tabletop into a tray, completing the task autonomously in real time (1x speed) throughout. Trajectories are generated by a fine-tuned VLA policy, with visual information from head and wrist cameras. The demonstration verified stable execution under drastic changes in ambient light, as well as the adaptability to re-recognize and continue completing the task after a human intervenes and moves target objects or the tray.

  3. Tokyo Robotics · YouTube38

    Torobo2 dual-arm robot demonstrates seamless task switching from laundry folding to table bussing

    Tokyo Robotics demonstrated its dual-arm robot Torobo2 seamlessly transitioning from "Laundry Folding" to "Table Bussing" in a continuous operation, completed autonomously in real time at 1x speed throughout. Trajectories are generated using a fine-tuned VLA policy, relying on visual information from head and wrist cameras to judge environmental changes. Each task has its own individual control policy (action model), and the policies are switched at appropriate timing to achieve uninterrupted continuous execution between tasks.

  4. Tokyo Robotics · YouTube40

    Torobo2 Autonomous Clothes-Folding Demo

    The dual-arm robot Torobo2 demonstrates autonomous clothes folding, using visual information from head and wrist cameras to generate trajectories in real time via a fine-tuned VLA policy, executing the entire process autonomously. The first half of the video completes the folding task at 1x speed, while the second half (played at 1.5x speed) shows its flexible adaptation and correct folding of clothing with different shapes and properties, such as shorts, a towel, and a T-shirt.

  5. EngineAI51

    EngineAI Unveils Embodied Intelligence Engine EngineAI Awaken at WRC, T800 Demonstrates Versatile Task Capabilities on Site

    EngineAI Robotics unveiled EngineAI Awaken, an embodied intelligence engine built on a five-layer human-brain-inspired architecture, at the World Robot Conference (WRC) opening on August 19, 2026, covering a capability chain from the S1 instinctive reflex layer to the S4+S5 self-growth layer.

Aug 20Thu
  1. 加速进化 · 微信公众号72

    Tsinghua's Zhao Mingguo team publishes Science Robotics paper proposing a unified perception-motion reinforcement learning framework for vision-driven humanoid soccer skills

    On August 19, Professor Zhao Mingguo's team in Tsinghua University's Department of Automation published the paper "Learning Vision-Driven Reactive Soccer Skills for Humanoid Robots" in Science Robotics, selected for the August 2026 humanoid robotics special issue; the robot platform used was a humanoid robot independently developed by Booster.

    Why it matters: The paper breaks down three failure points of the traditional perception-planning-control pipeline and presents a reproducible sim-to-real transfer method.

Aug 19Wed
  3. X Square Robot/自变量机器人45

    X Square Robot debuts at WRC: dual-scenario demos in home services and logistics sorting, with hands-on QUANXTA Zero bodyless data collection

    At the Beijing World Robot Conference (WRC), held August 19–23, X Square Robot showcased three exhibition areas: home services, logistics sorting, and bodyless data collection. Its fully self-developed end-to-end unified world model WALL-B drives robots to complete dozens of household chores. Logistics sorting reaches an efficiency of 1,816 items per hour with 98% accuracy, at a cost 70% lower than the "humanoid robot + five-fingered dexterous hand" approach, and has already entered routine operation on real production lines. The QUANXTA Zero data collection solution achieves a data-ingestion validity rate above 85% and can reduce training-data costs by 90%.

Aug 18Tue
  1. 荭杏科技28

    Hongxing Technology Dazhuang Z6 Industrial Humanoid Robot | Flexible Industrial Productivity That Can Think

    Hongxing Technology has launched the Dazhuang Z6 industrial humanoid robot, which uses a native integrated VLA architecture to connect perception, cognition, and decision execution, targeting smart factories, unmanned warehousing, precision assembly, and dynamic flexible production line scenarios. The Dazhuang Z6 adopts a dual-layer architecture of an intelligent brain plus a motion cerebellum, with large-model decisions directly driving millisecond-level motion control; it can read natural-language task instructions and acquire new skills by observing demonstrations, without complex programming or deployment. The company says that based on reinforcement learning plus imitation learning, it learns while working and supports minute-level line-change debugging on dynamic production lines.

  3. Skild AI76

    Skild AI releases robot foundation model S1, achieving in-context learning through video demonstrations

    Skild AI released the robot foundation model S1, which uses video demonstrations as task prompts and can execute manipulation tasks not seen during training without fine-tuning, with long-horizon tasks lasting up to about 10 minutes. At a pretraining data scale of 100,000 hours, the ICL model achieved a 66% success rate on unseen tasks, compared with 9% for the language-prompted VLA; a single demonstration was equivalent to about 380 post-training demonstrations, and the planting task took 11 minutes from recording the demonstration to autonomous execution. S1 was trained on NVIDIA AI infrastructure and has already been put into use with commercial partners.

    Why it matters: The original article provides comparison figures for ICL and VLA at the same data scale and a deployment timeline, allowing readers to assess the feasibility of the in-context learning approach for manipulation tasks.

Aug 17Mon
  2. 未来不远 · 微信公众号45

    Future Not Far F2 robot autonomously makes bolognese at 2026 WRC and demonstrates multi-robot household collaboration

    At the 2026 WRC World Robot Conference, the Future Not Far F2 robot autonomously made bolognese pasta on site through 7 consecutive steps with two-arm coordination and no human intervention, based on the second-generation fully self-developed body along with the new-generation WAM world action model and AVLA model. At the same event, the company publicly demonstrated multi-robot long-horizon closed-loop capability for the first time: one robot autonomously collected dirty laundry, loaded the washing machine, and managed the wash cycle, while another took over folding and stacking, all with no human involvement. Four new colors were also unveiled on site: Sweet Cherry Pink, Champagne Gold, Deep Sea Blue, and Mint Green. Booth C136, exhibition dates 8.19–8.23.

Aug 15Sat
  1. 灵巧智能23

    100% Pass Rate! 2026 Embodied Intelligence Summer Training Camp Concludes Successfully

    In August 2026, the 2026 Embodied Intelligence Summer Special Training and Study Camp, hosted by Zhejiang Lingqiao Intelligent Technology Co., Ltd. and co-hosted by Shanghai Artificial Intelligence Research Institute and Zongshen Chuangke (Shanghai) Technology Co., Ltd., concluded. Over the 5-day, 40-class-hour program, all trainees from more than ten institutions and dozens of enterprises and public institutions passed the assessment, with an overall satisfaction score of 96.8 points.

  2. Sergey Levine30

    Chelsea Finn on robots entering the GPT era

    Physical Intelligence co-founder Chelsea Finn spoke at Startup School 2026 about building general-purpose robots in the real world: reinforcement learning doubled robot throughput, and the system can run autonomously for hours. She believes robots are entering the "GPT era"—moving from specialized models to general systems that work across tasks and across robots. Sergey Levine remarked that Chelsea's talk was once again rock-star level, and that robots really can fold laundry and make espresso!

Aug 14Fri
  1. 国科投资 · 微信公众号57

    Light Origins Completes Pre-A Funding Round of Hundreds of Millions of Yuan, Led by CAS Investment

    Light Origins announced the completion of a Pre-A funding round of hundreds of millions of yuan, led by CAS Investment, with participation from China Merchants Venture, Xianghe Capital, and others. The funds will be used for large-scale embodied intelligence model training, multimodal data infrastructure development, full-stack software and hardware platform R&D, and team expansion.

  2. Harbin Institute of Technology Robotics65

    HIT's Wu Zhiguang team develops adaptive magnetically controlled nanorobot swarm platform, achieving navigation in the spinal canal of mice and non-human primates

    Professor Wu Zhiguang's team at Harbin Institute of Technology (HIT), together with collaborators, developed an adaptive magnetically controlled nanorobot platform that can form swarm motion patterns such as vortex, chain, and ribbon under external magnetic field drive, and achieved targeted navigation within the spinal canal in mice and non-human primates. The results were published in Science Advances.

    Why it matters: The research team released experimental results on navigation of magnetically controlled nanorobot swarms in non-human primates. Readers interested in the medical translation of micro- and nanorobots can learn about its technical approach.

Aug 13Thu
  2. Elite Robots/艾利特机器人64

    Elite Embodied Dual-Arm Robots Enter Apple Supply Chain Drilling Workshop for Mass-Production Loading and Unloading

    The embodied dual-arm loading and unloading robot jointly developed by Elite and Maixing Robotics has achieved 24/7 unmanned mass production in the drilling workshop of a leading Apple supply chain factory, with a single-cycle takt time of 52 seconds, a loading and unloading yield of 99.9%, and 5 units stably online. The solution uses hybrid force-position control and dynamic trajectory planning to achieve dual-arm coordination, is equipped with the self-developed Yuanqi Primo embodied large model, achieves loading accuracy within ±0.2 mm, and uses a wireless IO four-step handshake scheme to eliminate the need for modifications by the original machine tool manufacturer, shortening the deployment period from 10 days for the first unit to 7 days for the fifth unit.

    Why it matters: The original text provides actual performance figures such as line takt time, yield, and deployment period, allowing readers to understand the implementation path of the embodied dual-arm solution in a mass-production workshop.

Aug 11Tue
  1. 艾利特机器人 · 微信公众号65

    Elite embodied dual-arm robots enter Apple supply chain drilling workshops for mass-production loading and unloading

    An embodied dual-arm composite loading and unloading robot jointly developed by Elite and Maixing Robotics has been running in mass production in the drilling workshop of a leading Apple supply chain factory, with a single-cycle takt time of 52 seconds and a loading/unloading yield of 99.9%; the first unit has run continuously for more than a year, and 5 units are online simultaneously.

    Why it matters: The original text provides the takt time, yield, and deployment period for dual-arm loading and unloading in an Apple supply chain drilling workshop, which can be compared with single-arm and humanoid approaches to assess engineering trade-offs.

  2. 艾利特机器人 · 微信公众号40

    Elite Robots featured on CCTV2: "one brain, multiple forms" global smart manufacturing strategy makes it a Jiangsu going-global benchmark

    Elite Robots appeared on CCTV2's CCTV Finance. Its products have entered more than 50 countries and regions worldwide, exports accounted for 33% of sales in 2025, and overseas market sales grew about 170% in the first half of this year. In May this year, the company globally unveiled its wheeled humanoid robot Centaur-G1, which uses a "wheeled omnidirectional mobile base plus high-precision dual arms" configuration, and it self-developed an industrial-grade embodied large-model platform called "Yuanqi Primo." Its order schedule is currently booked through mid-next year, and it is shifting from an equipment supplier to a solutions provider.

  4. 正行创新 Striding AI · YouTube29

    How STEAM Enables High-Precision Autonomous Robot Unboxing | StridingAI Demo

    In a StridingAI demo, the STEAM model from Professor Chao Yu's team at Tsinghua University drives a robot to autonomously complete a multi-stage unboxing task, achieving millimeter-level bimanual coordination through tool use. STEAM models task progress and action quality frame by frame, extracting learning signals from mixed-quality data including expert demonstrations, model rollouts, and human takeovers to improve task success rates and execution speed.

  5. 达纳灵动 · X68

    Dyna Robotics releases Dyna-2 world-action model, pretrained on one million hours of human video

    Dyna Robotics introduces Dyna-2, a world-action model pretrained on one million hours of human video. The authors say they found several new scaling laws at this scale: world-action models across four orders of magnitude on human data (from 1,000 hours to 1.

    Why it matters: The original text provides data scales spanning four orders of magnitude and three scaling law conclusions, which can help readers understand how human video pretraining transfers to robot data.

Aug 9Sun
  1. LeadeRobot77

    Mila and Polytechnique Montréal team open-source ~$2,000 robot guide dog Milo, built on Unitree Go2 for autonomous indoor and outdoor navigation

    Mila-Quebec AI Institute and a Polytechnique Montréal team have released Milo, an open-source robot guide dog that they say can cut the roughly $50,000 cost of a guide dog to about $2,000. The hardware base is a Unitree Go2 Air/Pro plus an NVIDIA Jetson Orin Nano; a magnetic-encoder telescoping handle is used to inversely solve the user's 3D position, a bird's-eye-view map is built first and then a navigation policy is run, and the policy is trained in about 10 minutes in GPU-accelerated simulation and deployed zero-shot on the real robot.

    Why it matters: The original article fully breaks down Milo's perception, policy training, and real-world testing details, so readers can reuse its engineering trade-offs in their own navigation projects.

  2. LeadeRobot79

    Italy's IIT and INAIL unveil ergoCub humanoid robot; human-robot collaborative lifting can reduce load on the human lumbosacral joint

    The Italian Institute of Technology (IIT) and INAIL proposed a shared embodied intelligence architecture and built the humanoid robot ergoCub; the research was published in Nature Machine Intelligence.

    Why it matters: The original article lays out methods for building ergonomics into robot hardware optimization along with real-robot data, letting readers understand how design and control are organized around human-robot collaboration.

Aug 8Sat
Aug 7Fri
  1. UniX AI/优理奇 · YouTube17

    UniX AI's Panther humanoid robot autonomously prepares a family breakfast

    In its home breakfast service scenario, UniX AI's Panther humanoid robot autonomously plans and completes the full workflow, from preparing ingredients and heating food to making drinks and serving the meal at the table. UniCortex handles long-horizon task planning, UniFlex supports rapid learning of new household skills from a few demonstrations, and UniTouch combines visual and tactile perception to improve manipulation reliability for food, kitchen tools, and everyday objects. More information at www.unix-group.ai.

Aug 6Thu
  1. 澎湃新闻科技频道 · 微信公众号51

    Interview with Fudan's Chen Tao: Embodied foundation model architecture urgently needs a breakthrough; a GPT moment is unlikely within two years

    Chen Tao, director of Fudan University's Deep Learning Laboratory, said in an interview with The Paper's tech section that mainstream VLA models rely too heavily on successful-trajectory data and lack failure and error-correction data, leaving robots without autonomous error correction or continual learning capabilities. The embodied foundation model architecture urgently needs a breakthrough, and an embodied GPT moment will be hard to achieve within two years.

Aug 4Tue
  1. OFweek Robotics71

    X Square Robot open-sources HOST framework; robots can learn new skills from a 29-second video

    X Square Robot has released and open-sourced the HOST framework, in which a robot can reproduce a new skill after watching an average 29-second human demonstration video, achieving a 62% success rate with only 1/50 the data and 1/500 the learning time of the Pi-0.5 fine-tuning approach. HOST does not imitate human actions frame by frame; instead, it uses visual outcomes as an intermediate language, aligning video with robot execution by task progress rather than time. It adopts a dual-expert MoT architecture, with learning occurring at the inference stage while weights stay frozen, and it retains old skills better than the fine-tuning approach; under lighting and scene changes, the success rate remains above 50%. The paper and code have been open-sourced.

Aug 3Mon
Jul 31Fri
  1. RoboParty · 微信公众号67

    RoboParty Lab, Zhejiang University, and Tsinghua AIR release and open-source world model control method INTACT

    RoboParty Lab, Zhejiang University, Tsinghua University Institute for AI Industry Research (AIR), and InSpatio jointly released and open-sourced the world model control method INTACT (INtent-To-ACTion).

    Why it matters: The original article provides latency and success-rate comparisons with traditional CEM search, which readers can use to judge the benefit boundaries of search-free world model control.

  2. Ars Technica65

    Google releases Gemini Robotics 2.0, claims improved dexterity and safety

    Google has released Gemini Robotics 2.0; robots driven by the Gemini AI model can complete more complex tasks, continuously analyze changing environments, and collaborate with other robots. This version includes three new sub-models, among which the upgraded embodied reasoning model Gemini Robotics ER 2 is described by DeepMind as a significant improvement over the previous 1.6 version and has been integrated with the Gemini Live API; another sub-model is available to developers starting that day.