China has deployed robotic snakes to inspect power lines and help ensure a stable electricity supply.

The snake-like robots wrap around power lines and use onboard cameras and sensors to detect hazards such as damaged wires, worn components, and abnormal temperatures.

According to online media reports, the robots have inspected more than 80 miles (130 kilometers) of distribution lines serving exam sites near airport no-fly zones.

The system is three times more efficient than manual inspections, reducing the need for labor-intensive, hazardous work while improving reliability and safety.

Robotic grid guardians

Power utilities in southwestern China are using robotic snakes to inspect power lines, showcasing a new approach to infrastructure monitoring in environments where drones face limitations.

The snake-like robots have been deployed in Kunming, the capital of Yunnan province, where they move directly along power lines while inspecting critical electrical infrastructure. Equipped with cameras and sensors, the robots can identify faults such as broken wires, worn components, and abnormal temperature changes that could signal potential failures, reports the South China Morning Post (SCMP).

Developed by the Guandu District Power Supply Bureau, the robotic system has already inspected more than 80 miles (130 kilometers) of power distribution lines. According to its developers, the technology delivers inspection efficiency three times higher than traditional manual methods.

Unlike aerial drones, which are widely used for power-line inspections, the robotic snakes operate directly on the cables they monitor. This allows them to work in locations where drone flights are restricted, including areas near airport no-fly zones. The robots are also less vulnerable to high-voltage electromagnetic interference, a common challenge that can disrupt drone control systems and reduce image quality.

Meet the snake-like #robot patrolling power lines in #Kunming.

Slithering along cables, it uses onboard cameras and sensors to spot damage and overheating in real time. Even better, it draws power straight from the line, enabling autonomous 24/7 monitoring without interruption. pic.twitter.com/zXJq61oft4

— China SCIO (@chinascio) June 10, 2026

The system addresses several limitations of drone-based inspections, including restricted flight zones, weather dependence, and limited battery endurance. By providing continuous close-range monitoring of power infrastructure, the robotic snakes enable utility crews to quickly identify and address potential hazards, reports SCMP.

Intelligent grid inspectors

Chinese power utilities are deploying a range of robotic systems, including snake robots, robot dogs, drones, and surveillance cameras, to monitor electricity infrastructure and maintain reliable power supplies during the national college entrance exams.

The snake robot features a flexible, multi-jointed body that allows it to move along power lines and navigate around obstacles such as insulators. This design enables it to reach areas that are difficult for human inspectors and aerial drones to access. A camera and onboard sensors mounted in the robot’s head detect potential faults and safety risks, while a non-contact power-harvesting system allows the robot to draw energy directly from the power lines it is inspecting.

Following successful trials on multiple transmission routes, utility operators have expanded the robot’s deployment to more complex sections of the power grid. Robot dogs are also being used for inspections in high-voltage environments and other locations that are challenging or hazardous for human workers. Combined with drones and fixed monitoring systems, the robotic platforms provide continuous oversight of critical electrical infrastructure.

The deployment comes as approximately 12.9 million students sat for China’s National Higher Education Entrance Examination, known as the gaokao, one of the country’s most important academic tests. Reliable electricity is considered critical during the examination period, prompting utilities to adopt advanced robotic technologies to improve inspection efficiency and reduce operational risks, reports SCMP.

A four-legged robot dog designed to detect toxic substances could soon become the first responder sent into dangerous areas and help firefighting crews assess the damage without risking their own safety.

The remote-controlled detection robot was developed by researchers at the Graz University of Technology (TU Graz) in Austria. It is a compact, quadruped system that also incorporates measurement devices already used by fire services.

According to the research team, the detection robot has already been extensively tested by emergency services as part of large-scale pollutant drills. It has proved its ability to provide real-time data from danger zones.

Gerald Steinbauer-Wagner, PhD, an associate professor and robotics researcher at TU Graz, revealed that the dog can be deployed on missions ahead of firefighting crews. “It measures pollutant concentrations and delivers the data together with a camera image in real time from the danger zone,” Steinbauer-Wagner explained.

First into danger

In chemical accidents, emergency crews need to identify hazardous substances quickly while operating under dangerous conditions. Firefighters must typically don bulky protective suits, such as a Level A or B Hazmat suit, prior to entering contaminated areas to take measurements.

Although necessary to ensure total isolation from hazardous environments, the process can consume valuable time during an emergency. To address this issue, the team turned to robotics.

As per the researchers, the robotic dog can be sent in before firefighters to collect critical information from the scene. “This allows the emergency services to assess the situation without having to enter the danger zone themselves,” Steinbauer-Wagner continued.

The technology reduces firefighting teams’ direct exposure to toxic substances. It could significantly improve their safety during industrial accidents and hazardous materials incidents.

“Based on a comprehensive requirements analysis, we developed an appropriate robotic strategy and integrated it into existing standard operational procedures,” Steinbauer-Wagner added. “This should ensure seamless collaboration between man and machine during the operation.”

A safer perspective

Instead of creating a new system, the team made it compatible with existing fire-service procedures. The researchers conducted an extensive requirements analysis before integrating it into established hazardous materials response protocols. The robot relies on commercially available parts for mobility, sensing, data processing, communication, and visualization.

“The results of the field tests were extremely promising in terms of mobility, user-friendliness and effectiveness,” Steinbauer-Wagner reported. Meanwhile, a recent potential analysis supported by the Austrian Research Promotion Agency (FFG) highlighted several advantages of the technology.

Gerald Czech of the Austrian Federal Fire Service Association emphasized that, in the event of a real incident, he would send the robot dog into the contaminated area before the firefighters go in. “For the fire service, every minute counts during a hazardous materials operation,” he said in a statement.

Czech further said receiving measurement data and images from the danger zone while crews are still putting on protective gear could save valuable time during an emergency. “This increases the safety of the crew and improves operational command,” he concluded.

Experts from the Styrian and Upper Austrian fire service associations, TU Graz, the University of Applied Sciences Upper Austria, as well as the Disaster Competence Network Austria, support the use of detection robots. They propose making them a permanent part of fire departments’ hazardous materials response teams.

💾

Google has unveiled DiffusionGemma, a new experimental AI model that generates text using diffusion rather than the autoregressive approach used by most large language models today. The company says the model can deliver up to four times faster text generation on dedicated GPUs while running on consumer hardware.

The model builds on Google’s Gemma 4 family and Gemini Diffusion research. Unlike traditional language models that generate text one token at a time from left to right, DiffusionGemma creates and refines blocks of text in parallel.

According to Google, the approach enables output speeds exceeding 1,000 tokens per second on an NVIDIA H100 GPU and more than 700 tokens per second on an NVIDIA GeForce RTX 5090.

The company says DiffusionGemma is aimed at developers working on speed-sensitive applications such as interactive editing, rapid content iteration, code infilling, and other workflows where low latency is more important than maximum output quality.

Breaking text bottlenecks

Most large language models generate text sequentially, predicting one token after another. While effective, this process can leave local hardware underutilized when serving a single user.

DiffusionGemma takes a different approach. Instead of generating text word by word, it creates a 256-token block at once and then repeatedly refines it through multiple passes.

Google compares the difference to moving from a typewriter to a printing press. Rather than waiting for each token to be generated before producing the next one, the model processes an entire section of text simultaneously.

The company says this shifts the bottleneck from memory bandwidth to compute performance, allowing modern GPUs to operate more efficiently during local inference.

Another key feature is bi-directional attention. Since the model generates text in parallel, every token can attend to every other token during generation. This makes it better suited for tasks where future context matters, such as code completion, in-line editing, mathematical structures, and biological sequences.

Google highlighted a demonstration in which DiffusionGemma was fine-tuned to solve Sudoku puzzles, a task that can be challenging for conventional autoregressive models because later tokens influence earlier decisions.

Designed for local AI

The model uses a 26-billion-parameter mixture-of-experts architecture but activates only 3.8 billion parameters during inference. According to Google, this allows the model to fit within roughly 18 GB of VRAM when quantized, making it accessible on high-end consumer GPUs.

DiffusionGemma also includes an iterative self-correction mechanism. Because it evaluates an entire text block during refinement, it can identify and fix mistakes as generation progresses.

However, Google acknowledged that the model prioritizes speed over quality. The company said standard Gemma 4 models remain the preferred choice for production environments where output quality is the primary concern.

The speed advantage is also most apparent in local deployments and low-concurrency environments. In cloud settings serving large numbers of users simultaneously, conventional autoregressive models can often utilize hardware efficiently through batching, reducing the benefits of diffusion-based generation.

Google has released DiffusionGemma under an Apache 2.0 license through Hugging Face and is supporting deployment through tools including MLX, vLLM, Hugging Face Transformers, NVIDIA NeMo, and Unsloth.

Humanoid robots may soon gain more human-like dexterity thanks to a new wearable device developed by researchers at the Massachusetts Institute of Technology.

The ultrasound wristband uses high-frequency sound waves to capture the movement of muscles, tendons, and ligaments beneath the skin, creating detailed data on human hand motions.

According to the team, the data collected by the system could be used to train robots to perform household tasks by replicating the dexterous hand movements used by humans.

Last year, researchers at the University of California San Diego developed a wearable AI system that accurately interprets arm gestures and controls robots, even during running, vehicle motion, or rough ocean conditions.

Wristband teaches robots

For robots to perform delicate tasks with human-like precision, they must first understand the complex movements of the human hand. Researchers have now developed a wearable ultrasound wristband that could help bridge that gap by capturing detailed muscle and tendon activity beneath the skin and translating it into robotic motion.

Tracking hand and finger movements is a key challenge in fields such as robotics and spatial computing. Existing solutions have significant limitations. Camera-based systems can struggle with restricted viewing angles and visual obstructions, while wearable devices based on strain sensors, inertial sensors, or electromyography often limit movement or lack the resolution needed to accurately capture continuous finger motions, according to the team’s study paper.

The new wristband builds on a series of advances in wearable ultrasound technology over the past several years. Researchers have steadily expanded the capabilities of ultrasound wearables beyond gesture tracking and robotics. In 2022, scientists demonstrated a wearable ultrasound device capable of continuously monitoring multiple internal organs. Subsequent developments included wearable systems for continuous heart imaging and bladder volume monitoring.

In 2024, researchers introduced a conformal ultrasound patch capable of tracking cerebral blood flow in three dimensions and an integrated wearable ultrasound system that monitored deep tissue activity with sub-millimeter resolution using a single transducer. More recently, flexible, disposable ultrasound patches were developed to reduce manufacturing costs.

Now, in a new study, researchers led by Xuanhe Zhao at MIT introduced a wearable ultrasound imaging wristband capable of tracking arbitrary hand configurations. The device uses high-frequency sound waves to image muscles, tendons, and ligaments inside the wrist, providing a detailed view of the mechanisms that drive hand movement.

Decoding human hands

The wristband could help robots achieve human-like hand dexterity by capturing the movements of muscles and tendons beneath the skin. The device incorporates a 256-channel wireless ultrasound imaging system that uses high-frequency sound waves to monitor structures inside the wrist responsible for controlling finger and palm motions.

The ultrasound data is processed by a hybrid Transformer–ResNet AI model capable of interpreting subtle muscle activity. The system continuously tracks all 22 degrees of freedom of the human hand, covering the full range of joint movements that allow fingers and the palm to bend, rotate, and coordinate complex actions. Achieving this level of tracking has been a longstanding challenge for hand-motion sensing technologies, reported AP News.

In laboratory tests involving eight volunteers, the wristband reproduced hand gestures with high accuracy and low latency, mirroring movements within 120 milliseconds. Researchers demonstrated the system’s ability to recognize all 26 letters of American Sign Language, highlighting its precision in capturing intricate finger configurations.

Because the wristband operates wirelessly, users can control robotic systems remotely without being physically connected or located in the same room. To showcase its capabilities, the team used the technology to manipulate three-dimensional objects in virtual reality and direct a robotic hand to play the piano.

Beyond teleoperation, researchers believe the system could be used to generate large-scale datasets of human hand movements, providing valuable training data for future humanoid robots and helping them learn complex manipulation skills more autonomously.

In a landmark achievement for robotics, a bicycle-riding robot has performed the world’s first unassisted front flip, astonishing engineers and robotics enthusiasts alike. The feat, accomplished by a team at the University of Tokyo’s JSK Robotics Laboratory, marks a significant leap in the capabilities of autonomous machines and could have far-reaching implications for robotics research, mobility, and future transportation.

Engineering a robotic stunt rider

The robot, dubbed “Rider-Bot,” is built to mimic the movements and balance strategies of human cyclists. Unlike previous robotic stunts that required ramps, wires, or external supports, Rider-Bot executed the front flip entirely on its own—using only its onboard sensors, actuators, and algorithms.

The successful flip took place on a flat surface, with the robot accelerating, launching, rotating 360 degrees forward, and landing upright—all without human intervention.

According to the research team, achieving this level of acrobatic control required a combination of advanced gyroscopes, real-time feedback loops, and machine learning techniques. “This is a major step toward robots that can adapt dynamically to complex environments and perform tasks that were previously considered impossible for machines,” said Dr. Yuta Sato, lead researcher on the project.

Implications for robotics and beyond

The successful front flip demonstrates not only the mechanical resilience of Rider-Bot but also the sophistication of its AI-driven decision-making. The robot’s control system anticipates and compensates for shifts in momentum and balance, much like a skilled human cyclist would.

Researchers believe this breakthrough could inform the design of next-generation delivery robots, autonomous vehicles, and even disaster-response machines capable of navigating unpredictable terrain. “Our work shows that robots can be taught to handle extreme maneuvers, which opens up many possibilities for robust, adaptable mobility in real-world scenarios,” Dr. Sato noted.

Challenges and next steps

Despite the successful demonstration, the team acknowledges ongoing challenges. Perfecting the landing and ensuring consistency across different environments remain top priorities. The researchers plan to enhance the robot’s learning algorithms and test it in outdoor and obstacle-rich settings.

The team also hopes their achievement will inspire further research into dynamic locomotion and the integration of robotics into everyday life. “We’re just beginning to scratch the surface of what intelligent, mobile robots can do,” said Dr. Sato.

A milestone for human-robot collaboration

The Rider-Bot’s unassisted front flip not only pushes the boundaries of robotics but also highlights the growing intersection between human athleticism and machine intelligence. As robots become more capable of complex movement, their potential roles in society—from urban mobility to emergency response—are poised to expand dramatically.

With this record-breaking feat, the University of Tokyo’s robotics team has set a new benchmark for autonomous machines, paving the way for future innovations in robotic agility and intelligence.

💾

A new humanoid robot from China became one of the standout attractions at a major industrial exhibition after performing a Tai Chi routine.

Zoomlion’s Z01 bipedal robot demonstrated smooth, coordinated movements, highlighting its advanced motion control, dynamic balance, and human-like mobility.

The Tai Chi demonstration at KOMATEK 2026 in Istanbul showcased the robot’s ability to execute complex sequences with precision while maintaining stability throughout the performance.

In January 2026, the construction giant unveiled a full-chain AI system to drive intelligent transformation across machinery, manufacturing, management, and robotics.

Advanced balance display

While detailed specifications have not been released, Z01 is reported to feature millisecond-level motion response, long battery life, powerful onboard AI computing, and a high-degree-of-freedom design that enables smooth, natural movement. It also includes a built-in operating system that reportedly reduces robot development complexity by three to five times.

The robot’s Tai Chi routine showcased sophisticated motion control systems, coordinated full-body movement, and dynamic balance capabilities. By maintaining smooth transitions between poses and adapting its posture throughout the sequence, Z01 demonstrated the level of control required for real-world human-centered environments. Visitors closely followed the performance as the robot carried out the traditional martial art movements with remarkable consistency and coordination, according to a statement by Zoomlion.

Beyond the demonstration, Z01 represents a broader push into embodied AI and intelligent robotics. The humanoid robot is intended to support future applications that require close collaboration with people, including industrial assistance, guided services, and educational interaction. Its development reflects growing efforts to combine advanced software intelligence with robotic hardware capable of operating in dynamic environments.

“The debut of Z01 highlights Zoomlion’s growing commitment to embodied AI, with embodied AI robots and related emerging industries forming the company’s third growth curve,” said the firm in a statement.

AI robotics ecosystem

Zoomlion is advancing embodied AI through a technology stack that combines robotics hardware, intelligent software, motion control, and industrial automation. Building on years of investment in industrial IoT, the company has developed a strong digital infrastructure through Zvalley, its AI-powered industrial internet platform, supported by nearly 1,300 technical and R&D specialists.

Since 2024, the company has accelerated development of embodied intelligence systems by integrating robot hardware, core components, perception technologies, AI decision-making engines, motion-control algorithms, and software ecosystems into a unified framework. This approach enables robots to perceive their surroundings, make autonomous decisions, and execute complex physical tasks in dynamic environments.

By the end of 2025, Zoomlion had developed eight embodied AI robot prototypes spanning four categories, including humanoid and wheeled robotic platforms. These systems have been tested across real industrial settings, where they perform tasks such as logistics handling, factory inspections, loading and unloading operations, pre-assembly processes, and quality control. Operating in live production environments has allowed the company to gather valuable real-world data for training AI models and refining robot performance.

At Hannover Messe 2026, Zoomlion further demonstrated its technological progress through Robot Ops, an embodied AI development platform designed to coordinate intelligent robotic systems. The platform showcased autonomous decision-making, advanced motion control, multi-robot collaboration, and industrial deployment capabilities, highlighting the company’s efforts to transform embodied AI from research concepts into scalable industrial solutions.

💾

Anthropic has opened the doors to its most advanced AI system yet, but not without placing guardrails around its most sensitive capabilities.

The company on Tuesday unveiled Claude Fable 5, the first publicly accessible version of its Mythos-class AI model. Anthropic says the model outperforms every generally available system it has released to date, delivering major gains in software development, analytical work, visual understanding, and scientific reasoning.

The launch marks a significant shift for the AI firm. Until now, Mythos-level capabilities remained restricted to a small group of trusted organizations because of concerns that the technology could aid cyberattacks and other harmful activities. Fable 5 changes that equation by bringing those capabilities to a broader audience while routing certain high-risk requests through a less permissive model.

Safety before scale

Anthropic said Fable 5 includes a new layer of AI-powered classifiers designed to detect potentially dangerous prompts and jailbreak attempts. The company stress-tested those safeguards before releasing the model to the public.

If users submit requests involving cybersecurity, biology, chemistry, or model distillation, the system automatically redirects those interactions to Claude Opus 4.8 instead of Fable 5. Users receive a notification whenever that fallback occurs, allowing them to understand why the switch happened.

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.

Its capabilities exceed those of any model we’ve ever made generally available. pic.twitter.com/2AvmEjHIX8

— Claude (@claudeai) June 9, 2026

The company acknowledged that the safeguards may occasionally block harmless requests. However, Anthropic said early testing showed that fewer than 5% of sessions triggered a fallback response. “We’ve deliberately tuned the safeguards to be cautious,” the company said, adding that it plans to refine the system over time to reduce unnecessary interventions.

Built for demanding work

Anthropic argues that Fable 5 delivers its biggest advantages when tackling lengthy and complex assignments. The company says the model can stay focused across extended workflows while maintaining consistency over large amounts of context.

During early evaluations, payments company Stripe reported that the model completed a large-scale migration across a 50-million-line Ruby codebase in roughly one day. According to Anthropic, the same task would have otherwise occupied an engineering team for more than two months.

The company also highlighted gains in financial analysis. On Hebbia’s finance benchmark, Fable 5 posted the strongest results among leading models. Trading firm IMC found that the system performed well across factual analysis, reasoning tasks, root-cause investigations, and expected-value calculations.

Fable 5 also advanced Anthropic’s vision capabilities. The model can extract detailed information from scientific figures and reconstruct web applications from screenshots. In another demonstration, it completed Pokémon FireRed using a minimal vision-only setup after earlier Claude models struggled with the challenge.

Mythos stays restricted

Alongside Fable 5, Anthropic introduced Claude Mythos 5, a version of the same underlying model with fewer restrictions in select areas. Anthropic says the model offers the strongest cybersecurity capabilities it has developed so far.

Access to Mythos 5 remains limited. The company is deploying it through Project Glasswing while expanding availability to approved cyber defenders and organizations responsible for critical infrastructure. Anthropic said it intends to broaden access further through a trusted program in the future.

Fable 5 is available through Anthropic’s Claude API and consumption-based Enterprise offerings. Through June 22, Pro, Max, Team, and seat-based Enterprise subscribers can access the model at no additional cost. After that date, Anthropic plans to move users to a credit-based system before eventually restoring broader subscription access.

As competition intensifies across the AI industry, Anthropic is betting that powerful models do not have to arrive without safeguards. With Fable 5, the company is testing whether frontier capabilities and tighter controls can coexist in the hands of everyday users.

💾