AGI Bot X2 Webster Flip Meta AI Glasses AI Agents Protocol Persistent 3D World Gen AI

Just this May, the company launched a partner recruitment program with plans to move into large-scale production in the second half of this year, and they expect to sell and ship several thousand robots by the end of next year as humanoids are being used for broader applications, both at home and at work. That only leaves two questions. How much will the robot cost, and who owns the data it produces when it’s inside of your home or office? For comparison, the Unitree’s G1 costs about $16,000 and it’s also achieving shocking levels of agility, raising the question of why we need humanoids that are faster, smarter, and tougher than humans are.

Is this really in our best interest? More on that soon, but for now, the G1’s ability to recover already outperforms a human, and its hands are about to get a similar update, as Wooji Tech is making headlines with the debut of its first-generation AI-powered robot hand, which is engineered as a modular upgrade for general-purpose humanoids. In fact, the hand’s design focuses on three areas, with the first of these being its ultra-lightweight components. At its core, the hand’s skeletal structure weighs less than 600 grams for a slim and agile profile that’s physically sturdy enough for rugged tasks.

And to tie it together, the second design feature is the Wooji hand’s fully-actuated architecture with no coupled joints, meaning each finger is independently controlled, thanks to 20 rotary joints, all of which are individually powered by brushless servo motors. This configuration enables an impressive range of abduction movement, closely replicating the biomechanics of a human hand in a one-to-one scale form factor. But the third and possibly most important feature is power, with the hand being capable of delivering fingertip forces greater than 15 newtons, while its static grip loads exceed 20 kilograms. This allows the hand to perform tasks ranging from gentle manipulations to secure heavy lifts.

Plus, the addition of a soft glove around the hand’s skeleton increases its grasping ability for greater safety and smoother real-time adaptability when interacting with delicate objects. And beyond hardware, the Wooji hand leverages a high-speed communication framework, with all 20 axes communicating at 1000 Hz, with 20 output encoders and 20 input encoders providing real-time feedback for every movement. This ensures not only speed but also a precision of about 10 micrometers, placing it among the most precise robot hands currently available. And for potentially dangerous tasks, the hand supports full teleoperation, allowing it to be controlled from anywhere.

This could open up new possibilities for human operators by improving their fine motor skills to allow them to do almost anything from anywhere. Because of this, interacting from a distance may become the new norm with the inevitable rise of haptic feedback gloves to truly connect people without borders. But there’s another new piece of AI hardware that may transform our lives even sooner, as Meta just introduced its Ray-Ban display glasses. These smart glasses aim to blend advanced artificial intelligence features with everyday convenience, letting users interact with digital information while remaining engaged in the real world.

At the core of Meta’s Ray-Ban display glasses is a discrete, high-resolution, full-colour display positioned off to the side of the lens. This setup offers quick glance access to messages, photo previews, translations, real-time navigation and more, with the display remaining out of sight when not in use. It is designed for short, controlled interactions, allowing users to check updates or get help from Meta AI without the need to reach for their smartphones. On top of this, each pair comes paired with the Meta Neural Band, an electromyography wristband. This wearable translates subtle muscle signals from the user’s wrist into digital commands, providing a truly hands-free method to scroll, click, or even, in the near future, write out messages with small finger movements.

And the glasses don’t compromise on connectivity. With built-in microphones, speakers and cameras, users can privately view messages from services like WhatsApp, Messenger and Instagram, plus take live video calls from their contacts. Audio playback is also seamlessly integrated, with gesture controls for navigating tracks and adjusting volume directly from the Neural Band. The price starts at $799, so tell us in the comments whether you think this will make humans smarter or dumber over the next few years. And when it comes to achieving full-scale robot AGI, Figur just announced it closed its Series C funding round with over $1 billion of committed capital to solve General Robotics, reaching a post-money valuation of $39 billion.

But tell us in the comments below when you think they’ll start selling Figur 3, and if the company goes public, would you buy their stock? Simultaneously in France, a new contender in the industrial robotics market is taking shape with the debut of the Calvin 40 humanoid. Developed in just 40 days, this human-sized robot leverages its platform that’s been refined over 12 years with billions of simulated and real-world steps. As a result, Calvin 40 now features an impressive level of lower limb durability after having performed more than 1 million steps. It also comes equipped with adaptable end-effectors to make it customisable for a range of demanding or ergonomically hazardous tasks.

Plus, Wondercraft says the robot is running medical-grade software to ensure operator safety while its feet have integrated force sensors for precise centre-of-mass control and rapid balance adjustments. As for the robot’s environmental awareness, a day-and-night camera system allows the Calvin 40 to manipulate objects, navigate spaces and avoid obstacles in real-time. And to let people know what the robot is up to at any given time, it has LED light strips that communicate its status with nearby collaborators. Moreover, because of its computer vision and reasoning, Calvin 40 can quickly deploy across different workflows to tackle real-world manufacturing tasks.

But the digital world is also transforming with AI as Marble just launched its beta for persistent 3D world creation. For a peek into the future, this limited-access beta preview lets users create, view and navigate persistent 3D worlds on their browser. This is a result of the company’s ongoing expansion on the frontier of spatial intelligence with a focus on delivering navigable controllable environments that remain consistent, regardless of how long or how deeply they’re explored. With Marble, users can generate expansive three-dimensional worlds using either text prompts or images as input. The resulting environments are notably larger while featuring greater stylistic diversity and improved geometric cleanliness compared to previous iterations.

Plus, unlike some earlier generative models that imposed session time limits or experienced mid-session morphing and inconsistencies, Marble’s worlds are persistently stable and free from such constraints. What’s more is that Marble enables creators and enthusiasts to export their generated environments as Gaussian splats, which is a kind of efficient representation for three-dimensional spatial data. These exports integrate seamlessly with downstream projects through the open-source Spark rendering library, which supports rendering within 3JS across a range of devices, including desktops, laptops, mobile devices, and virtual reality headsets. Furthermore, users can also compose multiple individual generations to construct large, stylistically coherent worlds, marking a significant leap towards scalable virtual environment design.

Looking into the future of applications like the Metaverse, the model’s consistency and adherence to stylistic cues pave the way for a series of creative applications from game design to web-based interactive experiences. It’s important to note, though, that the current version of the Marble model excels in creating immersive environments but does not yet specialise in generating isolated central objects such as people or animals. However, the model’s flexibility with input styles ranging from flat cartoons to highly realistic imagery empowers creators to iterate freely and find the look that best fits their project needs.

And beyond the 3D world, what if AI agents could pay and be paid, forming an AI economy where the whole value of the network is greater than the sum of its parts? Well, it’s finally happening as Google just released the Agent Payments Protocol, also known as AP2, which is a new open standard designed to allow AI agents to securely conduct payments across various platforms. And by building on top of the foundation of the existing protocol, AP2 finally combines a range of payment methods, including major credit cards, stablecoins, as well as traditional bank transfers.

But an even bigger feature of AP2 is its introduction of digital mandates that act as cryptographic authorisations to capture and lock in user intent. These mandates ensure that all transactions, whether happening in real time or through automated processes without direct user oversight, remain both verifiable and secure. Amazingly, this initiative has already gained the support of more than 60 of the most influential companies, including Mastercard, PayPal, Coinbase and Adobe. Furthermore, by establishing a unified framework, Google’s Agent Payments Protocol can create a trustworthy, standardised system for the agentic economy.

And in an effort to encourage adoption and transparency, Google has published its full documentation for AP2 on GitHub. And finally, a new hands-on workshop is making it easier than ever for developers and tech enthusiasts to build powerful AI report generation agents using NVIDIA’s Neotron NanoV2 model and OpenRouter’s API endpoints. The workshop walks participants through each step, starting with clicking the Deploy Now button. Afterwards, users are led to a self-contained, launchable environment where all required resources and dependencies are already installed. Plus, the setup process is streamlined. Just add your API keys through the built-in Secrets Manager and you’re ready to roll.

Once deployed, the environment offers direct access to JupyterLab, including extra tools like an introduction to agents and the Secrets Manager button for seamless key management. Furthermore, the course covers defining search tools, assembling researcher agents, and building conditional workflows that automate everything from research to report writing. On top of this, the approach is fully guided, making it an ideal entry point for anyone looking to automate report writing with state-of-the-art AI tools. Anyways, like and subscribe for more AI news and check out this video here. [tr:trw].