And it’s built with 55 degrees of freedom, including its ultra dexterous hands, which have demonstrated a series of real-world tasks around the house. Furthermore, the robot features not only the ability to transfer skills from human demonstration to robot execution, but also can be remotely operated, allowing anyone to carry out any physical task with near human ability. This is because the robot’s movement is driven by a tendon-based system to provide even greater control over the force it applies. And with the company’s upcoming 2025 release of the robot to the market, its price tag is supposed to cost about the same as a mid-range car between $15,000 to $40,000.

But that’s only the beginning as Magic Bots also just released more footage of its fully electric robots powered by advanced joint motor modules delivering up to 525 Nm of peak torque. This impressive combination of power and precision allows these robots to perform an astonishing variety of tasks. From handling delicate objects like fragile glassware to heavy lifting operations, the robots demonstrate a versatility that is reshaping industrial workflows. But unlike other traditional hydraulic systems, Magic Bots’ compact electric drive system is energy-efficient and sustainable, making these humanoids more practical for deployment in a variety of industrial settings. But the real hallmark feature of Magic Bots lies in their highly dexterous hands, with each one being equipped with six mini-high torque servo actuators and precision pressure sensors, enabling submillimeter accuracy.

Additionally, its hands allow the robots to mimic up to 70% of human hand gestures, making them suitable for intricate operations like circuit board assembly, quality control, and packaging delicate items. In fact, this level of precision has opened up new possibilities for automating previously labor-intensive tasks, offering manufacturers a competitive edge in efficiency and output quality. But these early showcases are just a glimpse of the company’s broader future vision, with Magic Lab currently testing its humanoids on an electronics factory assembly line, as this pilot project is expected to refine the robot’s capabilities for mass deployment across multiple industries.

And Magic Lab is also in partnership with ByteDance, the parent company of TikTok, to integrate the Dubow Large Language Model into its third-generation Magic Bot. In fact, the Dubow model is more cost-effective compared to models like OpenAI’s GPT-4, but still offers a range of capabilities, including generating text, creating images, and powering advanced video creation tools. And this Dubow integration could evolve beyond mechanical tasks to handle creative and communicative functions, such as generating visual content, composing operational reports, or even interacting with human colleagues in a seamless, conversational manner. Magic Lab’s advancements come as part of a broader push by China to mass-produce humanoid robots by 2025, aiming to dominate the global market by 2027.

But as for an AI breakthrough that’s available right now, Google DeepMind just released Genie 2, a new foundation world model that generates endless playable 3D environments, plus unlocks the next tech paradigm for training and evaluating AI agents. At its core, Genie 2 addresses a long-standing challenge in AI, the need for diverse, complex, and scalable environments to train embodied agents, AI systems that operate in and interact with the physical world. With Genie 2, researchers can now generate practically unlimited scenarios, from bustling cityscapes to tranquil forests, all from a single image prompt, and the applications go far beyond AI training.

Genie 2 could revolutionize game development, virtual prototyping, and even creative workflows for designers. Impressively, Genie 2 is a world model system that simulates virtual environments, including the consequences of actions such as jumping, swimming, or interacting with objects. Trained on a massive video dataset, Genie 2 uses state-of-the-art AI techniques, such as autoregressive latent diffusion models, to generate rich and realistic 3D environments. By combining this with powerful text-to-image generation tools like Imogen 3, Genie 2 can transform a simple text prompt into an interactive world. Moreover, Genie 2 can generate a range of possible environments based on any description, then after selecting a preferred rendering, users or AI agents can step into the world and explore it through keyboard and mouse inputs.

Plus, Genie 2 is already demonstrating a range of advanced capabilities, starting with intelligently responding to user inputs, such as moving characters with arrow keys or interacting with objects. It understands context, for example, knowing that arrow keys should move a robot character and not a tree or cloud in the scene. And starting from the same initial frame, Genie 2 can even generate multiple distinct trajectories based on different actions taken. This feature allows researchers to simulate what-if scenarios, providing diverse experiences for training agents. But what’s truly incredible is that unlike other models, Genie 2 even remembers parts of the world that are no longer in view.

For example, if an agent walks behind a house and returns, the system accurately renders the house as it appeared before. Additionally, Genie 2 can also generate new content on the fly while maintaining consistent worlds for extended periods. It supports various perspectives, including first-person, isometric, and third-person views. From bursting balloons to realistic water and smoke effects, Genie 2 models complex interactions and physical behaviours. It simulates gravity, lighting, reflections, and even the subtle motion of grass in the wind or water flowing in a river. On top of this, Genie 2 can animate diverse characters, from humans to animals to fantastical creatures, performing a wide range of actions.

Moreover, the model includes interactive NPCs, enabling more complex simulations involving multiple agents at once, and the potential applications for Genie 2 extend far beyond AI research. In fact, Genie 2’s diverse environments allow researchers to create novel tasks and scenarios for embodied AI agents. For example, researchers can test an agent’s ability to follow natural language instructions in unseen environments. Designers can use Genie 2 to rapidly prototype new game worlds and mechanics. Imagine creating an entire level by simply sketching a concept or describing it in words. Artists can transform concept art or rough drafts into fully interactive environments, accelerating the creative process for virtual worlds.

Most of all, it could serve as the backbone of the coming generative gaming revolution. Anyways, like and subscribe and watch more Genie 2 bonus footage. Thanks for watching!
[tr:trw].