Techman Robots TM Xplore 1 Unveils New NVIDIA AI Powered Humanoid Game Changer

➡ Techman Robot has introduced a new humanoid robot, TMXplore-1, with advanced movement, sensing, and learning capabilities, making it ideal for use in warehouses and production lines. Meanwhile, WeRobotics has showcased a robot named Alex with exceptional dexterity and strength, but its cost and release date are yet to be announced. In other news, Quenimage has launched Quenimage Edit, an image editing model that allows users to modify images while maintaining their original intent and visual consistency. Lastly, Dynamics Lab has unveiled Mirage 2, a generative game world engine that lets users create and modify interactive worlds in real time.

Today on AI News, Techman Robot just unveiled a new humanoid named the TMXplore-1 with NVIDIA onboard. But what are its specs? As for physical design, the robot features 22 axes of movement for human-like dexterity. Next, in terms of perception, TMXplore-1 features full 360-degree sensing capabilities and environmental recognition for ultra-precise positioning, making the robot well-suited for use in warehouses and production lines to start. And when it comes to intelligence, the robot inherits its dexterity using the company’s proprietary TMAI vision system. And for grasping, TMXplore-1 uses Techman Robot’s in-house developed motor modules, plus a sophisticated control architecture for reliable performance in sorting and assembly.

Additionally, thanks to the TMVision language action model, the robot is trained using both teleoperation and simulation data, generated on NVIDIA’s Isaac platform. By leveraging tools like NVIDIA’s Isaac Groot Mimic and GrootGen, the robot can generate large-scale training trajectories, allowing it to better interpret natural language prompts. This powers the Cosmos Predict system, which can visualize up to 30 seconds of continuous video, helping operators plan and preview robot actions. Plus, the TMXplore-1 is designed to work alongside other collaborative robots to extend its utility across various logistics and manufacturing scenarios. And the robot features both flexible lifting and intelligent balancing, using its high-precision wheelbase for stable yet agile movement across dynamic workspaces.

But for superhuman dexterity, WeRobotics from Korea just released another video showing its Alex robot carrying out incredibly delicate tasks with extremely precise force control requirements, demonstrating finger repeatability of 0.3mm. In fact, Alex features 15 degrees of freedom in its hand alone, giving it a level of dexterity that may even exceed that of an average human. For instance, Alex can screw things around the house using just its thumb, and its fingers are all back-driveable with ultra-low friction while remaining compliant with less than 100g of kinesthetic force sensing. On top of this, Alex features a lifting capacity of up to 30kg per arm and a fingertip force of over 40N.

In fact, Alex has 48 degrees of freedom throughout its entire body, with 7 per arm, 15 per hand, plus neck and waist articulation, allowing Alex to be ultra-responsive in any setting. But when it comes to the robot’s cost and release date, this information still hasn’t been disclosed by WeRobotics, so make sure to tell us in the comments how much you’d be willing to pay. And for superhuman walking, figure just released more footage of its 0.2 humanoid walking across all kinds of uncertain terrains, including uneven grass, scraps of wood, metal, and even glass, all without losing its footing.

This is thanks to the robot’s Helix AI system that has been trained using reinforcement learning to handle changing environments with even more grace than most humans. The 0.2 humanoid’s advanced sensors and real-time processing allow it to analyze and respond to surface changes instantly, and in some areas it’s already starting to reach superhuman levels, with the robot walking blind, meaning without the use of any cameras, throughout the demonstration. As for the figure 0.3 launch date, the company has tentatively planned alpha testing in homes for the end of 2025. But tell us if you would trust this humanoid in your home alone, and should it be sold on a subscription basis like a cell phone? And beyond embodied AI in humanoids, in the latest generative AI news, the team behind Quenimage has officially introduced Quenimage Edit, a new image editing model that builds upon their existing 20 billion parameter Quenimage platform.

This latest development not only extends Quenimage’s advanced text rendering skills, but also incorporates a wide range of image editing capabilities, aiming to bring both precision and flexibility to digital content creation. In fact, at the heart of Quenimage Edit is its dual-pathway architecture. When a user provides an image, the model simultaneously sends it to Quen2.5 VL for semantic visual control, as well as a variational autoencoder for managing visual appearance. This parallel processing enables the system to tackle both high-level semantic editing and meticulous appearance editing. What it means is that now anyone can alter an image’s content, rotate objects or transfer styles, all while preserving the original intent and visual consistency.

But another standout feature of Quenimage Edit is its proficiency in semantic editing. Put simply, semantic editing allows users to modify the core content of an image, such as generating new intellectual property, rotating objects through a full 180-degree range and transforming the overall style without losing the underlying meaning or character. A compelling example from the Quenimage Edit showcase is the transformation of a mascot. While the majority of the image’s pixels may change, the mascot’s identity and emotional tone remain consistent, demonstrating the model’s nuanced grasp of visual semantics. Plus, this capability opens doors for novel applications like generating a series of themed emoji packs, such as those inspired by the 16 Myers-Briggs Type Indicator personality types, which can effortlessly expand a brand’s visual language.

On top of this, the model excels in appearance editing, a process focused on adding, removing or modifying elements within an image while leaving the remaining areas untouched. Imagine inserting a new signboard into a photo and instantly seeing a realistic reflection appear or changing a subject’s background or clothing with no distortion to the rest of the composition. Quenimage Edit’s attention to detail ensures that such edits blend seamlessly, supporting practical needs ranging from marketing collateral to digital avatars. Furthermore, Quenimage Edit distinguishes itself with its precise text editing functionality, supporting both Chinese and English languages.

Users can even add, delete or adjust text within images, all while maintaining the original font, size and style. This extends to complex challenges like modifying intricate headline fonts or tiny annotation text in posters. The technology’s robust text rendering foundation, inherited from the original Quenimage model, empowers highly accurate and visually coherent edits. Benchmarking results further underscore Quenimage Edit’s industry-leading performance. Evaluations across multiple public image editing benchmarks have confirmed that the model achieves state-of-the-art results, cementing its position as a reliable foundation for advanced image editing tasks. And in generative AI gaming, Dynamics Lab has introduced the Mirage 2 demo showcasing its newest generative game world engine.

Now users can transform their own sketches and photos into interactive worlds, plus modify these worlds in real time simply by typing commands. On top of this, creations can be saved and shared with others. Mirage 2 also redefines gaming with its UGC 2.0 framework, enabling players to shape worlds dynamically through text, keyboard or controller inputs. Unlike traditional games with fixed narratives and environments, Mirage 2 allows real-time creation. Players can summon a misty forest, spawn a hovercraft, or erect a towering skyscraper mid-gameplay, with the engine seamlessly integrating these elements. Its photorealistic visuals surpass the blocky aesthetics of earlier systems, delivering lush landscapes and detailed urban scenes.

Mirage 2 supports extended play, with interactive sequences lasting over 10 minutes, offering deeper engagement. However, demos reveal limitations, such as occasional lag in processing complex inputs and minor visual flickering during rapid world changes. Anyways, like and subscribe to AI News and check out this video here. [tr:trw].