Skild AI BRAIN Omni Robot Control (ANY ROBOT ANY TASK)

Skilled AI has developed a new AI model, Skilled Brain, that can control any type of robot and perform any task. It uses a two-layer system to translate commands into precise movements, and it learns from a huge amount of data, including human behavior. The company has also created a wristband that lets people control computers with their muscle signals. Meanwhile, a UK company has improved its robot hardware, and a new robot, the Galaxy R1, is being launched that can move like a human and handle heavy loads.

Is this the beginning of robot AGI? Skilled AI just announced breakthrough results in artificial general intelligence research with the introduction of Skilled Brain, a singular, omnibodied foundation model designed to work across any robot type and for any task. Unlike previous approaches that typically can find AI control to specific hardware or form factors, the Skilled Brain is engineered to operate seamlessly across a wide spectrum of morphologies. This includes quadrupeds, humanoid robots, tabletop arms, and mobile manipulators. Plus, intriguingly, even incorporates data from human behavior on the premise that humans themselves can be viewed as a form of robot.

But the key feature of the Skilled Brain is its hierarchical architecture where at its top level, it uses a high-level action policy that handles navigation and manipulation commands at a low frequency, while it also employs another lower-level policy that translates the low-frequency commands into precise joint motions and motor talks in real time. This dual-layered approach is fine-tuned through a combination of large-scale simulated experience and targeted real-world data, meaning the robot performs steadily even as hardware or environments change. On top of this, Skilled AI addresses the challenge of data scarcity in robotics by leveraging trillions of simulated and internet video-based training examples, far exceeding the data scale typically seen before.

This massive data set, together with real-world fine-tuning, allows the Skilled Brain to generalize across tasks beyond the simple pick-and-place operations where many models plateau. Furthermore, the model is deployed via an application programming interface, abstracting away low-level control details and enabling developers to focus on high-level applications. But robots are only the beginning, as researchers have introduced a non-invasive neuromotor interface that enables users to control computers through their muscle signals, all without the need for invasive surgery. Using a highly sensitive, easy-to-wear wristband equipped with surface electromyography, or SEMG sensors, the system captures electrical activity from the muscles.

Plus, a vast infrastructure was built to collect training data from thousands of consenting participants, which made it possible to develop machine learning models that are not limited to a single individual. Furthermore, users could handwrite at speeds up to 20.9 words per minute. On top of this, personalizing the SEMG decoding models improved handwriting performance by 16%. This marks the first time a high-bandwidth neuromotor interface has delivered strong, out-of-the-box generalization across a broad population, opening doors for more natural computer input methods in everyday scenarios. But robots can only perform as well as their actuators physically allow them to, which is why humanoid’s latest announcement about validating its newest hardware is so important.

In fact, the UK-based company just integrated dual seven-degree-of-freedom robotic arms onto its base platform, and early results hit all key benchmarks. Precision kinematic calibration was confirmed using a one-millimeter sphere test, and the robot’s digital twin model mirrors its real-world motion accurately. Mechanical and assembly quality checks have also all passed, plus cable routing and strain relief, meaning the company is well on its way to enabling expanded functionality, as well as broader deployment opportunities moving forward. Meanwhile, a brand new contender is entering the home robotic scene, the Galaxy R1, aiming to set a new benchmark for intelligent, human-like assistance in real-world environments.

As for its specs, the R1 features a full-body design with 24 independent degrees of freedom, enabling highly flexible and fluid movement. Each of its dual robotic arms features seven axes of motion, closely mirroring the dexterity and range of the human arm. These arms together can handle a maximum payload of 10 kilograms, reaching across a span of 64 centimeters, from ground level to as high as 2 meters. And for even greater precision, the arms are equipped with force sensors and support a variety of end-effectors, including dexterous hands for intricate tasks. But the R1 also introduces advanced interaction capabilities with multi-dimensional perception, with its standard configuration integrating seven high-definition cameras and a laser radar sensor so the robot can navigate and understand complex environments.

The R1 even supports voice interaction for easier usage. As for raw computation, its architecture is built on the powerful NVIDIA Jetson AGX Auron module with 32 gigabytes of memory, offering a computing power of 200 trillion operations per second. This helps to drive the 6 degree of freedom vector chassis, delivering true 360 degree omnidirectional movement at speeds up to 1.5 meters per second. Furthermore, the R1 is designed for both isomorphic and remote operation, offering VR compatibility for telepresence tasks too. And in the latest generative AI news, OpenAI’s ChatGPT just got a new upgrade called Study Mode, and it’s designed to tackle tough homework, prepare for exams, and dig into new topics.

Importantly though, this new dedicated feature has been crafted in partnership with teachers, scientists, and learning experts. Because of this, Study Mode guides students through interactive prompts, scaffolded explanations, and knowledge checks that are all designed to deepen comprehension rather than simply provide answers. Plus, the tool adapts to each student’s skill level, organizing information into manageable sections with personalized support and feedback. And Study Mode offers flexibility too, letting users toggle the feature on or off depending on their learning objectives. Experts say Study Mode is a positive step toward using AI to foster curiosity, critical thinking, and real engagement throughout the learning process.

But the competition is heating up, as Google’s Notebook LM just got a major boost with the introduction of video overviews and a redesigned studio panel. To start, video overviews bring narrated slide presentations to Notebook LM, offering users a dynamic way to understand complex topics. This feature combines newly generated visuals with relevant images, diagrams, quotes, and data pulled directly from users’ documents. Plus, users can customize these videos to focus on specific subjects, set learning goals, and tailor content for different audiences, all in English for now, with additional languages coming soon. And in the Studio panel of Notebook LM, it now allows multiple outputs of the same type, such as several audio overviews or mind maps, all in a single notebook, making it easier to organize material for different teams, audiences, or study purposes.

In fact, users can even multitask within the studio, like listening to an audio overview while viewing a mind map. And finally, Vibe Coding just got a major upgrade thanks to the release of Microsoft’s GitHub Spark, its newest AI-powered app creation tool that’s now available in public preview and included with Copilot Pro Plus plans. What’s more is that Spark will also roll out to additional GitHub Copilot plans, with Microsoft stating that Spark aims to close the gap between idea and execution by weaving artificial intelligence directly into the software development lifecycle. But what’s special about it is that GitHub Spark finally enables anyone to build applications, not just with natural language prompts, but also using visual controls, targeted edits, and even traditional code.

With this new fine-grained power, Spark lets users turn simple descriptions into fully functional apps that feature both front-end and back-end support, all powered by Claude Sonnet 4. And there’s absolutely no setup required because everything from data handling to deployments is managed out of the box. Users can even add extra AI features using models from OpenAI, Meta, DeepSeek, and XAI, with no API key management necessary, and to make it truly accessible to even the non-technical crowd, Spark features one-click deployments and synchronised repositories to make development even smoother, while Copilot Agent integration supports rapid iteration.

Anyways, like and subscribe for more AI news and check out this video here. [tr:trw].