Aaron Ames, director of Caltech CAST, believes safety is the most critical yet missing element before humanoid robots enter society, and that the lack of clear safety standards will inevitably lead to injuries; he notes that current demonstrations are mostly performance-driven, and few companies are seriously considering safe deployment.
At an IEEE RAS Communications Committee meeting, several experts informally discussed the boundary between the concepts of "robotics" and "automation." Key points included: automation focuses on making processes repeatable, efficient, and scalable, while robots are its physical embodiment; however, not all robots are automated (e.g., fully teleoperated robots), and not all automation requires robots. Participants noted that vague definitions can affect national statistical standards for service robots, international comparisons, educational examples, and the evolution of terms such as "physical AI" along with research trends.
In an interview, Yuansheng Intelligence founder Yang Sicheng proposed that when model capabilities are not strong enough, hardware capabilities must be very strong, and only after models break through can dexterous hands simplify. Its self-developed Apex Hand uses an in-palm hybrid tendon-driven scheme, has 21 degrees of freedom, a maximum single-hand active payload of 30 kg, and an open-close time of about 0.2 seconds, and was carried by the gold, silver, and bronze medal teams in the heavyweight categories of weightlifting and tug-of-war at the 2nd World Humanoid Robot Games in August 2026. He emphasized prioritizing reliability and dexterity, using unidirectional tendon drive plus end encoders to compensate for creep, and believes the endgame for dexterous hands is a reliable general-purpose productivity tool.
Leiphone visited vision-based tactile vendors including Daimon, Yimu, Weitai, Qianjue, and Moliang, reviewing the principles, product specifications, and deployment scenarios of this technology route and responding to doubts such as "a camera plus silicone is easy to replicate." The article notes that domestic vendors have reached frame rates in the hundreds of Hz, thicknesses in the 3 mm range, and that multiple products offer IP65 or higher protection; Daimon proposed the VTLA architecture and open-sourced 10,000 hours of tactile data, but cost, consistency and durability, and how to integrate tactile data into model training remain hurdles the industry has yet to clear.
Tomorrow in San Francisco: join our CTO and co-founder Wen-Loong Ma at the @runwayml AI Summit, speaking on how world models are reshaping #robotics and #PhysicalAI. ➡️ Location: The Masonic, 9 AM to 6 PM. Lineup and final tickets: https://summit.runway.com #robotics #automation #manufacturing @NVIDIARobotics
Unitree Robotics showcased the GD01 manned mecha at the 5th Global Digital Trade Expo. It is about 3 meters tall, weighs about 500 kilograms with a passenger, can switch between bipedal and quadrupedal modes, and is priced from 3.9 million yuan. Staff said there have been many inquiries but no sales yet.
Sentdex proposes naming the robot "Chonk" and asks for everyone's opinions. The main post calls it very cute.
@TryPallet regained its confidence during its hardest stretch. @SushanthRaman spoke with BCV partner @kevinzhang about what it took to get there, and how far the will to survive can take you. https://youtu.be/sM_L9zC1UYw?si=qEMDXzyGVWEJXYcC
Professor Zhang Tao, Dean of Tsinghua University's Institute for Embodied Intelligence and Robotics, delivered a lecture titled "Embodied Intelligence Empowers Everything" at the Tsinghua Humanities Forum. He described three major characteristics of embodied intelligence—generality, interactivity, and autonomy—and noted that key technologies still need to be breakthroughs, including multimodal embodied large models, environmental perception, dynamic planning and control, and dexterous bodies.
Hello Robot co-founder and CEO Aaron Edsinger introduces the company's Stretch robot, which is aimed at people with mobility issues and helps them live more independently. He also shares his earlier experience as a robotics director at Google and his journey through four startups, developing human-collaborative robots for the consumer, manufacturing, healthcare, and research markets.
A new blog post co-written with @perryadong discussing what robots need to be broadly useful in the real world. https://pd-perry.github.io/posts/post-training.html Reliability is one of the biggest unsolved challenges in AI today. Current models do okay when a human reviews their output (such as drafting code), but this will become a bigger bottleneck when we want systems to operate with greater autonomy and greater trust.
Ken Goldberg of UC Berkeley argues in an article for the Agentic Robotics (AR) approach, in which multi-agent systems write, test, and iterate structured robot control programs offline, as a third path between model engineering and data-driven methods that does not rely on large-scale robot demonstration data.
On September 8, KEENON Robotics COO Wan Bin joined UBTECH's Pang Jianxin and AgiBot's Yao Maoqing on stage at the inaugural Fortune Leaders Forum, held at the Grand Lisboa Palace Resort Macau, to discuss the commercialization and data challenges of embodied intelligence. Citing the deployment at Shangri-La Traders Hotel as an example, Wan Bin introduced the "general-purpose + specialized" service model, in which the XMAN-R1 humanoid service robot works alongside specialized robots such as the T10, T3, W3, C40, and S100.
What other failure modes exist? Quoted tweet key points: The blogger is friendly toward Figure but points out that its success rate is 237/420 (under 60%), and that failing half the time cannot count as "doing real useful work." Doing useful work = generalization + reliability.
Agentic Robotics — where AI agents design and fine-tune robot control systems — is being applied in real-world scenarios to improve e-commerce parcel sorting.
Tutor has released its next-generation robot, Sonny, a humanoid mobile robot equipped with the Ti series VLA built specifically for real-world scenarios, aiming to push the boundaries of robot capabilities in real customer deployments. Union Square Ventures commented that Tutor's R&D team is extremely fast, and even more impressive is the speed at which it deploys robots in the field—no demos, just real work in the real world.
Highpower Technology CTO Liao Xingqun detailed the technical logic behind the three growth curves of AI edge devices, AIDC, and robotics under the "AI edge battery + solid-state" strategy. The next-generation solid-state battery will be released in May 2026, with a volumetric energy density of 1200Wh/L and a gravimetric energy density of 400Wh/kg, about 30% higher than mainstream mass-production levels in the industry; it has so far only completed small-scale trial validation, while 950Wh/L solid-liquid batteries are already in mass supply. Robot batteries have been stably mass-produced and delivered to AgiBot, Zhishen, and UBTech, with related revenue expected to approach 100 million yuan in 2026.
This article is excerpted from the roundtable "Physical Intelligence: Divergences and Choices" at the 2026 Inclusion·Bund Conference, moderated by Koji of Crossroads, with Han Zheng of Sudo Technology, Shen Yujun of Ant Lingbo, Wang Qian of X Square Robot, and Xu Huazhe of PoKe Robot participating. The discussion centered on three sets of divergences: data sources (simulation versus real machines), whether Astra will deliver a crushing blow to embodied AI, and what metrics can prove that embodied AI has moved past the bubble.
In an interview, Xspark AI co-founder and chief scientist Ding Wenbo proposed that vision has already solved about 99% of robot perception problems, and that touch should form a low-latency, body-adjacent "spinal cord"-level model rather than being directly integrated into vision large models. He proposed a three-layer architecture of "slow brain VLM—fast brain VTLA+TWAM—spinal cord VTA," arguing that touch mainly serves collision-avoidance safety and post-training for dexterous manipulation, and noted that current bottlenecks lie in sensor consistency, tactile data representation, and unification across sensors, modalities, and embodiments.
Care facility leaders in every market we serve tell us the same thing: there isn't enough time in a shift to meet every need. Before starting development, we spent two years listening to these voices. Let's talk: https://enchanted.pulse.ly/qbho72ssnb #robotics #care #healthtech
Reliable whole-body data is more than a mesh that looks human. It requires precise ground truth, temporally consistent motion, and a production pipeline built for scale.
Speaking at the 2026 Annual Fund of Funds Forum and the 7th Lujiang Venture Capital Forum, Huako Fund founder Zhang Yang said super returns come from investing in "cognitive value" before technological value reaches consensus. He breaks returns into roughly 10x in the 0-to-1 stage, 5x to 10x in the 1-to-100 stage, and 2x to 3x after the 100 stage; multiplied together, these yield 100x. Huako invested in Zhipu in 2021 and added to its position across five consecutive rounds; Zhipu's valuation rose from several billion yuan to more than 20 billion yuan within a year. In the next stage, Huako will focus on quantum computing, controllable nuclear fusion, and robotics.
Never let robots perform comedy.
Foundation Robotics released the video "Ranking Iconic Robots" on YouTube, ranking iconic robots. The video's caption says "AI is always biased," and the music used is "Russian Dance" recorded by PM Music.
The 12th session of Fudan University's Intelligent Robotics and Advanced Manufacturing Lecture Series invited Su Hao, Haoqing Distinguished Professor and inaugural dean of the Institute of General Physical Intelligence, to give an academic talk. Nearly 500 faculty and students from on and off campus attended, and the event was held in a combined online and offline format for the first time.
During the World Robot Conference, Yunqi Capital co-hosted a themed roundtable with SEE Fund, which was included in the podcast Attent!on. Three founders—Xu Huazhe of PokeBot, Li Qiang of Quantum Dynamics, and Liu Nianqiu of Feimo Yi Technology—shared their views on world models, scaling, and deployment.
Why it matters: The three entrepreneurs approached the topic from models, hardware systems, and mobility respectively, specifically breaking down where the bottlenecks in embodied intelligence deployment lie and where they diverge on scaling efficiency.
Li Xiangming, general manager of RoboTechnik's Embodied Intelligence Business Unit, gave a keynote titled "How Industrial Robots Can Break Through in the AI Era" at the 2026 Global Industrial Internet Conference held in Liaoning, proposing four breakthrough paths: reliability, ease of use, full-category plus embodied intelligence, and going global.
At the close of an IEEE RAS Communications Committee meeting, several IEEE experts held an informal reflection on the terms "robotics" and "automation": automation makes processes repeatable, efficient, and predictable, while robots come into play when automation requires a physical presence. The experts noted that not all robots involve automation—some can be fully teleoperated by a person—and that not all automation involves robots, and can even be done purely in software. Differences in how the terms are defined can also affect cross-country statistics for products such as robot vacuum cleaners, perceptions of technological strength, and the examples available in education.
On August 29, Monolith held a closed-door brain-computer interface salon with various parties, summarizing several observations on where brain-computer technology stands and what it still lacks for a product leap. The article notes that technology routes have diverged into invasive electrodes, EEG, brain surface, vascular intervention, electronic skin, ultrasound, and others, and the market has not yet converged; brain data faces a data wall where quality, scale, and diversity are difficult to achieve simultaneously, and although public EEG data has accumulated to tens of TB, metadata is not standardized, and the industry has no consensus on whether to first solve acquisition or first understand existing signals.
The author argues that the future of home robots will not be a humanoid butler that does everything, but rather an AI robot brain coordinating the autonomous devices already in the home. At CES, she challenged every humanoid robot she encountered with a bottle of laundry detergent, and not one could open the cap.
OFweek published an analysis of the industry logic behind tactile sensor company PaXini's valuation crossing the 10 billion yuan threshold. After its Series B funding in March 2026, its valuation exceeded 10 billion yuan; in August it secured another 1 billion yuan in strategic financing, bringing cumulative funding to nearly 4 billion yuan. Over the past year, cumulative shipments of its self-developed tactile sensing chips surpassed 1 million units, and the GEN4 FUSE chip is based on four generations of architecture iteration.
Mark DiLeo, lead technician on Boston Dynamics' Atlas service and repair team, shares how he repairs the bipedal humanoid robot. He reveals that the worst case was when the robot fell and its leg completely split from its torso; the research version R1 requires one to two days of downtime each time it falls, while the production version D1 was designed for manufacturing and on-site customer repair, making it easier to service, and the team can usually repair it within one to two hours, getting the robot back to performing backflips.
Thanks to Ryan for visiting; this was a fun conversation. Key points from the quoted tweet: This episode of Peterman Pod features Sergey Levine, a top global robotics researcher and co-founder of Physical Intelligence, discussing the current state of humanoid robots, how China's robotics ecosystem compares with that of the United States, the possibility of OpenAI and Anthropic entering the robotics field, robotics research papers he recommends, and his predictions for when humanoid robots will be deployed.
At the Beijing robot sports event, someone filmed a small robot breakdancing. Luming Robotics replied with a like and said NIX's dance moves are very cool😎
Cheerleading first, then breaking—NIX is clearly here to steal the show. 🤣
The author reviews three dialogues hosted jointly by Leju and Ant Lingbo at WRC, covering whether embodiments are easy to learn, whether data is easy to use, and whether models are easy to evaluate, with 12 guests in total. Key viewpoints include that hardware is becoming a precondition for model capability, that Ant Lingbo's real-world data exceeds 100,000 hours and covers more than 20 configurations, that Noitom says about 20% of data plays a core role in model behavior, and that evaluation should guide adjustments to data distribution; on this basis, the author argues that industry attention is shifting from making a single movement to doing a task reliably and consistently.
Legend Capital has achieved 9 IPOs since the start of the year; Zhipu's market value briefly surpassed 1 trillion yuan, and ChangXin Technology's market value briefly surpassed 4 trillion yuan. The Social Security Zhongguancun Special Fund it manages (5 billion yuan in its first phase) has seen its first batch of portfolio companies—Zhipu AI, SJ Semiconductor, and Jingfeng Medical—go public. The fund has leveraged more than 30 billion yuan in social capital cumulatively, a leverage ratio of more than 9x.
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.
The World Robot Conference 2026 showcased a new signal: the focus of robot competition is shifting from "how human-like it looks" to solving real problems. This year's conference had nearly 50,000 square meters of exhibition space, with 373 companies presenting 3,000 innovative products and 311 new product debuts. Non-traditional form-factor products for special environments such as nuclear power operations and maintenance, mining, and rescue became the focus, including a 4-meter-tall hybrid hydraulic-electric super-humanoid robot and a centaur wheel-legged robot. The article argues that industry competition will shift toward cost, reliability, and data accumulation, and that future value will depend on the combination of hardware, AI models, data closed loops, and commercial scenarios.
ATDev has updated progress on its autonomous wheelchair development, building a modern power wheelchair equipped with a collaborative robotic arm under the ARPA-H-funded RAMMP project. The wheelchair uses a shared autonomy model in which the user decides the path and the system handles fine adjustments, with plans to add teleoperation and federated learning features. ATDev is working with Meta to apply AI models to the wheelchair, with both parties betting on open source.