Robotics news & what's new on Robolist

Companionship first or work first? The commercial fork in the road for home robots | Going global insights

How can embodied robots enter homes, and what sustains their ongoing operation? Qiyuan's choices reflect unresolved questions in this industry.

Geekplus Expands LPP Logistics Partnership to 1,600 Robots in Poland

Robotics and automation? Informal reflections on familiar terms

By Emmet Cole “Robotics and automation” is a pairing heard often around these parts. It puts the “RA” in IEEE RAS, after all. The familiar combination shows up in conference titles, funding calls, and program descriptions. But what is actually meant by the two words? Is robotics a subdomain of automation? Does the distinction matter? […]

Testing Qwen-3.8 vs GLM-5.3 Flash versions: which one lets me leave 30 minutes early?

On August 26, 2026, on the same day, Qwen released Qwen3.8-Flash-Next/Qwen3.8-Flash, and Z.AI released GLM-5.3-Flash. From their names, "Flash models" represent low price, speed, and handling high-frequency calls—more of a complement to flagship models. However, the official marketing focus from both companies this time has actually moved beyond cheap chat. (Leiphone) Qianwen Cloud positioned Qwen3.8-Flash directly toward 1M long context, long documents, complete code repositories, complex dialogue, programming assistance, workflow orchestration, and visual understanding, emphasizing its integration with developer tools; Zhipu's description of GLM-5.3-Flash goes further, incorporating visual programming, office tasks, financial research, document processing, and deliverables like PPTX, PDF, DOCX, XLSX into core selling points. Therefore, both models convey a new signal: Flash is not merely a low-cost complement to flagship models, but is entering the foundational layer of real productivity scenarios in office work, development, and deliverable generation. Thus, a more practical question emerges: if Flash models are designed to handle users' daily high-frequency tasks, what should they first prove? Our assessment is: models must first demonstrate they understand office work. Combined with the models' exceptionally strong comprehensive capabilities, we believe understanding office work should mean being able to comprehend…

From Hyrule to the Real World: How Video Models Understand 'Change'

HiDream-O1-Video's Three Global Response Tests. Author | Reah Editor | Li Na In The Legend of Zelda: Breath of the Wild, when a player ignites a patch of grass, flames spread along the blades, and hot air rises. At this moment, deploying the paraglider allows Link to catch the updraft and leave the ground. In a thunderstorm, equipping metal weapons may attract lightning, and rain makes climbing difficult. Hyrule does not pursue photorealistic visuals, yet through a stable, composable set of rules, it lets players experience a world in continuous operation. The key to this experience is that no matter how many types of objects exist in the world, the system responds to the player's actions in accordance with physical law. With stable rules, players can make predictions based on experience; when multiple rules interact, they produce solutions never explicitly programmed. A credible virtual world thus contains state, change, interaction, and feedback. Material source: Douyin creator 'Chou Chou Yu - Console Games' Zelda, of course, is not a world model in the machine learning sense. Its rules are constructed explicitly by designers and engineers; the latter attempts to learn the relationships between state, action, and consequence from data and interaction. The mechanisms differ, but both point to the same intuitive experience: a world works not because it looks real, but because it responds to actions in ways consistent with our physical world. This is also an effective entry point for understanding world models. A world model is not simply a larger-scale video generator, nor does it line up images, videos, three-dimensional space, and robots on the same matrix chart. It must form a representation of world states and their laws of change, and further support prediction, planning, or action. Zhixiang (HiDream.ai) has newly released HiDream-O1-Video-1.0 (hereinafter referred to as

Just now: GPT-6 Astra hooked up with Unitree G1, tidied the kitchen!

Let GPT invoke robot skills as tools

We Let GPT-6 Astra, Fable 5.1, and Sol Control the Same Robot: Which One Actually Got the Job Done?

Real-world gimbal test: three models all made the robot move, but their standards for 'completion' are worlds apart. Author | Rhea Editor | Li Na, Cen Feng. A gimbal robotic arm is a camera base/mount that rotates up, down, left, and right. I used three top-tier models—GPT-6 Astra, Claude Fable 5.1, and GPT-5.6 Sol—to make the robot move. All three completed the task, but their execution paths, validation standards, and costs differed significantly. After GPT-6 Astra's release, many excellent reviews have already appeared online. If readers can find those elsewhere, what new insights can this evaluation offer? In a purely digital environment, a mistyped line of code can usually be undone, and a misclicked button can be tried again. But the real world is not like that: physical devices have inertia, boundaries, and feedback—and real risks. Sending a command does not guarantee the action will be precise; what path to take, how fast to move, when to stop, and whether to verify the result all become choices the model must make itself. [First, let's watch a little dance.] A gimbal robotic arm is a perfect test case. Its motions are simple enough to understand without robotics expertise, yet speed, path, amplitude, and pauses are displayed directly before your eyes. Judgments hidden inside the model become visibly observable movement. (Spoiler: through the gimbal test, you can see at a glance the differences between the three models.) The author previously obtained a second-hand gimbal robot. The funny part is that the vendor kept the Raspberry Pi, so I had to use a Mac mini to drive it. When I started debugging, I discovered the robot's firmware had major issues—just bringing it back to life two months earlier burned through nearly 2,000 yuan in tokens.

How ‘Industry 5.0’ integrates human expertise with artificial intelligence

[The content of this article has been produced by our advertising partner.] The European Commission defines Industry 5.0 as placing human-centricity, sustainability and resilience at the core of industrial progress. In practice, this means AI-powered robots take on repetitive, physically demanding or hazardous work while people focus on critical thinking and higher-value decisions. The future of manufacturing is about using intelligent technology to help people be better at what they do, whil…

SFA, Rainbow Robotics join to advance AI autonomous manufacturing in Korea - CHOSUNBIZ

36 Robot Companies That Have Gone Public: Massive Differences in Profitability, No More Flashy Commercialization

Robots Go Public: The Wind Is Shifting

SFA Teams With Rainbow Robotics on AI Autonomous Manufacturing

Neubility Unveils Semi-Humanoid Robot 'Billy,' Targets Hundreds of Units Within Two Years

Texas A&M students build 50-pound drone that lifts 110 pounds

A team of Texas A&M University aerospace engineering students has built a 50-pound drone capable...

Lianhe Aircraft Qualifies for 2026 China Post Drone Procurement Project

On September 30th, according to United Aircraft, the shortlist of candidate suppliers for the 2026 China Post Unmanned Aircraft Equipment Model Resource Pool Procurement Project (Phase 2) was recently announced. United Aircraft Group (Beijing Zhonghang Zhi) was selected as a candidate supplier for the unmanned helicopter package. The core model awarded in this bid is the TD550 unmanned helicopter, which features compact and modular design, allowing rapid deployment without requiring large professional airports.

Tesla secures $30B in new credit lines as it looks to scale Cybercab, Optimus

The company says it won't draw on the new debt facilities this year, as it has already planned at least $25 billion in capital expenditures.

Would you pay for this robot to clean your house?

Can a robot clean your house better than a human?

MIT Engineers Design Swimming Robot Powered By Living Muscle Cells

China sets world record with 116 underwater drones navigating with zero human input

China has set a new world record for simultaneous autonomous underwater vehicles after 116 small...

AI boom drives robot makers Fanuc, Yaskawa to step up shift from China to US

Viral Jinsang Robotics crab robot "Jumper" goes open source, 100 units to be given away for free

IDC: Global humanoid robots enter explosive growth phase, with nearly 25,000 units shipped in first half, China's market share exceeds 70%

DH Robotics showcases multiple dexterous hands at IROS, world premiere of fully direct-drive 13-DOF dexterous hand

Mark Cuban Doubles Down On Criticism Of Humanoid Robots, Says Personal Robots Will Be ‘Optimized’

After losing his voice to cancer, this founder is building ‘glasses for voice’

Uhura Bionics, part of the Startup Battlefield 200 at TechCrunch Disrupt, wants to replace flat, robotics voice devices with one that carries emotions.

Leaked images reveal new colors for Amazon’s next entry-level Kindle

Amazon typically introduces new Kindle models a few months ahead of the holiday shopping season, and this year it looks like that will include a bigger update to the entry-level Kindle than what debuted in 2024. The last upgrades to the basic Kindle model were limited to a brighter backlight, a new dark mode, and […]

Meet Flourish One, the Raspberry Pi-powered humanoid built for busy parents

Powered by muscle cells, a paper-thin robot swims through a watery maze

Swimming can take a lot of muscle. But as MIT engineers have found, even a single layer of muscle cells can power through water if designed right. In a paper published Sept. 28 in the journal Advanced Functional Materials, the team presents a design for a thin, muscle-powered swimming robot. The "skeleton" of the aquabot is made from a film of gel about the length and width of a stick of gum. The two halves of the gel form the bot's "fins." Each fin is covered with a layer of live muscle cell…

Venus flytrap-inspired 'robot hand' material senses objects and grasps them under light

A Venus flytrap closes its leaves to trap prey when it senses a stimulus. Taking inspiration from this plant, KAIST researchers have combined the roles of skin that senses approaching objects and muscle that grasps them in a single soft material. When an electrically charged object comes near, ions inside the material move and produce an electrical signal.

MIT-led team tests wind turbines in air 220 times denser for higher power output

Researchers have used a highly pressurized wind tunnel to study how wind turbine alignment and...

MIT builds tiny paper-thin robot that swims using living muscle cells

After 1,000 Units Sold Out, Recreated as 'Han Li's Model': How Yuandian Robotics Brings Embodied AI Into Homes

What do users do with a humanoid robot after they buy one? Recently, Yuandian Xiaoqiao robot units have been gradually handed over to select media outlets and influencers for real-world testing and evaluation in actual use scenarios. Bilibili top creator 'Turing's Cat' offered a rather surprising answer: he used the Yuandian Xiaoqiao robot to recreate the 'Han Li-style puppet' from Riyao's Cultivation Biography. After the video went live, it trended on Bilibili's hot search, and the official account for Riyao's Cultivation Biography immediately 'liked' it in the comments. For Xiaoqiao, this was just another highlight in a string of achievements. After its official launch on September 2, the compact humanoid robot's first-day sales exceeded 600 units at the launch price of 8,888 yuan, topping Jingdong's humanoid robot sales chart; in less than a week, all 1,000 launch-batch units were sold out. Meanwhile, at IFA 2026 held in Germany, the Yuandian Xiaoqiao robot won the IFA Innovation Award in the 'Next & Emerging Technologies' category. According to IFA's official introduction, this award recognizes breakthrough innovations, prototypes, and early-stage technologies that point toward the future of consumer technology. From market sales and international awards to creative community adaptations, in just a few weeks Xiaoqiao claimed the spotlight in three completely different circles. But treating it as merely a suddenly viral humanoid robot may underestimate the wave of attention. Tracing the timeline further back reveals that Yuandian Robotics had previously launched multiple products around home embodied AI and achieved real market validation. Xiaoqiao resembles an especially outstanding representative within this suite of products and technologies. Xiaoqiao: More Than Just a Ten-Thousand-Yuan High-Performance Humanoid Robot Xiaoqiao is first and foremost a robot that is easy to remember. At 88cm tall and weighing approximately 13kg overall, with a peak torque of 50N·m

Hundreds of robots are roaming this Chinese theme park. It's only the beginning of our humanoid future

HDC Hyundai Development Expands AI Home Services With Health Care, Robots

Nvidia Alumni Are Fueling a New Wave of Robotics and AI Startups

Robotics startups are courting university researchers. Microagi just hired eight.

Narwal Flow 2 Flagship Robot Vacuum Gets Early Prime Day Deal

DyeusBot CEO Liu Chenhao: The Layered Revolution Behind Humanoid Robots' 40-Rally Badminton Exchange | IROS 2026

Latest Industrial-Grade Progress in High-Dynamic Movement for Humanoid Robots at Top-Tier Conferences. Author | Deng Zhemin Editor | Qi Chengyong Badminton might be the most 'unreasonable' of all ball sports. When a shuttlecock is smashed, its initial velocity can reach two to three hundred kilometers per hour, yet due to the enormous air resistance created by sixteen feathers, its speed decays sharply during flight. The trajectory is no longer a parabola—it becomes an asymmetrical arc with a straight front section and a steep descent at the back. Moreover, with cork at one end and feathers at the other, the center of mass and pressure center are offset, so each shuttlecock tumbles in mid-air of its own accord. This rotation makes its trajectory even harder to predict. In essence, playing badminton is a dynamic meditation: the body must move with agility while the mind stays calm enough to 'see' the landing point one second ahead. For humanoid robots, this combination of dynamic and static challenges requires a well-organized cerebellum-and-cerebrum system. The cerebellum layer has evolved rapidly over the past three years. Humanoid robot whole-body control has progressed from 'walking steadily' to 'moving like humans': first train robots to track various movements using human motion-capture data, then use system identification and domain randomization to reduce the gap between simulation and reality. Skills like running, jumping, and dancing have gradually stabilized, and work has emerged enabling robots to reproduce high-dynamic full-body movements, such as ASAP's method for aligning simulation with real-world physics. Against this backdrop, ball sports have become a new testing ground for cerebellum-and-cerebrum integration. Looking back at past work in the field, two trends are evident. First, layered architecture: AdaPT's core idea is to let the planner generate stylized movements while the tracker executes them with minimal interference; HITTER and LATENT also follow a 'high-level decision, low-level execution' structure. Second, human priors: motion capture, video, and amateur data all provide robots with experience of human movement. Yet the industry still faces

FF's Robotics Subsidiary FF EAI Robotics Implements Approx. 25% Equity Incentive Plan

China accounts for 77.9% of global humanoid robot shipments in H1, IDC says

Neubility Unveils Semi-Humanoid Robot 'Billy' in Push for Fast Commercialization

China has three new criteria for humanoid robot IPOs. Few, if any, meet them

Robot Tactile Technology War: Tens of Billions in Capital Floods the Sector in Half a Year, Multiple Companies Vie for the Title of "First Listed Tactile Technology Stock"

Roborock Tops Global Robot Vacuum Shipments in Q2 With 23.7% Share

From Creative Concept to Startup Launch, JLC Extends AI Hardware Innovation Support Chain

On September 19th, guided by the Futian District Science and Technology and Industry Information Bureau and hosted by JLC Group, the 4th Open Source Hardware Spark Conference and Global AI Hardware Innovation Ecosystem Conference was held at the Shenzhen Convention and Exhibition Center (Futian). This year's conference collected over 400 project submissions across open-source and closed-source tracks, covering embodied AI, robot key hardware, AI edge computing, intelligent terminals and other directions, with over 2,200 innovators from multiple countries and regions attending. The day before, the 2026 Spark Global Hardware Innovation Competition, targeting commercializable projects, held its final at Futian New Generation Industrial Park, with 20 teams from China and abroad competing after preliminary selection and accelerator empowerment, ultimately producing 9 award-winning projects. At the same venue, JLC's Open Objects Incubator officially launched. The conference released a new round of hardware innovation support measures covering R&D facilities, industry scenarios and market channels, and capital support. For a company that already provides design tools, component procurement, and manufacturing services, what does launching an incubator mean? The answer requires looking at the starting point of hardware innovation. A product concept needs to go through circuit design, component selection, prototyping, and assembly to become a functioning object ready for testing. Whether developers can easily access these services affects both the speed of idea realization and how many opportunities they have to iterate and modify. JLC's existing advantages are concentrated in this process from idea to prototype. Now, with the incubator launch, this support chain is extending further: helping developers bring ideas to life, while also...

Roborock tops global robot vacuum shipments as market premiumizes - CHOSUNBIZ

China's Social Logistics Total Reached 243.6 Trillion Yuan in First Eight Months, Up 4.9% Year-on-Year

The China Federation of Logistics and Purchasing announced logistics performance data for January–August today (September 29). The data shows that in January–August of this year, China's logistics sector maintained steady-progress development momentum, the logistics market expanded steadily, and international logistics corridors expanded and upgraded in capacity and quality. In January–August this year, total national social logistics reached 243.6 trillion yuan, a year-on-year increase of 4.9%. The steady growth in total social logistics volume and structural optimization proceed in parallel, and supply chain resilience in industrial chains has further strengthened. Specifically, the industrial goods logistics foundation remained stable, with industrial goods logistics volume in January–August growing 5.3% year-on-year, while demand for upgraded industrial manufacturing logistics accelerated. In August, high-tech manufacturing and digital product manufacturing added value grew 16.7% and 15.7% respectively, contributing over 60% to growth in industrial output above scale, with industrial robot production growing 34.6%, jointly driving new logistics demand.

Robotic Dogs Are Taking Different Forms As Their Mission Roles Expand

Flash news | HiMA Luxeed RX launches with starting price of 259,800 yuan, Stelato V8 opens pre-orders

On September 28, it was learned that the Luxeed RX officially launched, with the clear cabin roof lidar version priced from 259,800 yuan and the L3-level autonomous driving architecture version priced from 319,800 yuan.