➡ Three companies, Robot Era, Engine AI, and Clone Robotics, are making strides in the field of robotics. Robot Era’s X-Hand, attached to their X-Bot-L, has advanced features like sensitive sensors and powerful motor joints. Engine AI’s SE01 humanoid robot, priced at $5,300, is flexible and mimics human-like movements. Clone Robotics is developing human-like robotic hands that can grasp various objects. In the realm of 3D content creation, DreamCraft 3D++ from China and Adobe’s large-view synthesis model are revolutionizing the process by increasing speed and quality.

The stories you’re about to see will shock you. Firstly, Robot Era just previewed their newest X-Hand for embodied AI, and it’s attached to their full-size humanoid robot called the X-Bot-L. As for dexterity, the X-Hand demos its end-to-end visual neural networks for generalization and adaptation of various collaborative robot arms. Plus, tactile sensors on each of the fingers makes manipulation even more delicate by using high-precision array 3D force and temperature information. All of this technology uses on-chip calculation for real-time display of data while supporting EtherCAT to enable high-speed multi-concurrent real-time control. Furthermore, each finger joint is now both back-driveable and impact-resistant, meaning they are less likely to be damaged while working, and the X-Hand has 12 active degrees of freedom in each hand to set new standards in general-purpose robotics through four key features.

First, its fingertips contain highly sensitive array sensors that gather detailed information about the size, shape, texture, softness, and force feedback of objects, allowing for real-time grip adjustments and precise handling. Second, its simple yet effective transmission structure ensures smooth and reliable movements, minimizing mechanical failures and extending the hand’s lifespan. Third, the powerful motor joints can exert a maximum force of 80 newtons, surpassing human capability and enabling the X-Hand to easily lift heavy items like a five kilogram bottle of water. Lastly, the X-Hand seamlessly integrates with advanced AI models and algorithms, providing an ideal platform for future advancements in embodied AI and paving the way for more sophisticated robotic systems.

But there’s an even newer player in the robotics industry called Engine AI, and they’ve just revealed their SE01 humanoid robot with several human-like bionic features. But what can it do, and how much will it cost? To start, the SE01 stands out with its exceptional flexibility, boasting 32 degrees of freedom throughout its body. This feature enables it to perform complex and challenging movements with ease, thanks to a maximum joint torque of up to 330 Newton meters. And the SE01 is engineered to mimic a natural gait and human-like walking. By integrating a self-developed harmonic joint module with advanced reinforcement and imitation learning techniques, Engine AI is also pushing for a new benchmark in robotic mobility with the SE01’s end-to-end neural network models contributing to its smoothness and agility.

And as for its price, at $5,300, the SE01 is intended to be accessible and also open source, encouraging further innovation and development. With this design, Engine AI intends to make its technology widely available and cost-effective, including an expandable chest cavity and a multi-module quick-release interface for its head and hands, providing adaptability for various applications. In fact, the robot’s walking speed reaches up to two meters per second while achieving a battery life of approximately two hours. Plus, the robot’s full aluminum exoskeleton is durable in harsh conditions featuring integrated sealing and heat dissipation technology to maintain optimal performance.

And the SE01 humanoid overcomes the stereotypical unnatural gait of its predecessors with smooth and graceful walking as part of Engine AI’s broader mission to accelerate innovation in the area of embodied intelligence. In fact, a team of 36 researchers developed the SE01 using NVIDIA’s Isaac Jim, a robust simulation platform that allows for extensive virtual testing before real-world implementation. Importantly, Engine AI’s approach to full-stack self-research and the adoption of cutting-edge technologies both domestically and internationally have been crucial to this achievement. But that’s just the beginning because Clone Robotics is also developing cutting-edge general-purpose humanoids with a whole new approach and design, with the introduction of a pioneering bimanual torso equipped with an actuated elbow, cervical spine, and anthropomorphic shoulders.

In fact, this design incorporates realistic sternoclavicular, acromioclavicular, scapulothoracic, and glenohumeral joints with a compact valve matrix that’s neatly housed within the rib cage to enhance the robot’s bimanual manipulation capabilities. Moreover, the company’s efforts have also resulted in the creation of highly realistic robotic hands. These hands feature hydrostatic muscles under transparent skin, closely mimicking human appendages, with the V-15 biomimetic hand being able to grasp a variety of objects from tennis balls to suitcases. Importantly, the thumb and fingers are designed to replicate human appearance and function by using McKibben muscles, a mesh of tubes with internal balloons that contract via pneumatic or hydraulic pumps.

But Clone Robotics aims to minimize external components by focusing on electronically stimulated muscles instead, which the company achieves by filling balloons with acetaldehyde and applying electric currents to control muscle contraction efficiently. What’s more is that this process operates effectively at temperatures as low as 68 degrees Fahrenheit, with significant pressure increases at higher temperatures. Furthermore, the robotic hand features a skeleton with human-like bones and hinge joints, allowing mobility similar to a human hand. This offers approximately 27 degrees of freedom, with integrated hand and forearm rotations by using a network of tendons and muscles to control each movement, which extends through the forearm and hand.

As for power, Clone Robotics uses a basic hydraulic setup activated by a 500-watt pump at 145 pounds per square inch, distributing pressure through 36 electrohydraulic valves. Magnetic sensors provide data to onboard AI, which adjusts joint angles and velocities. And looking ahead, Clone Robotics wants to develop this technology and integrate it with their locomotion platform with support for a battery pack too. Meanwhile, in a breakthrough for 3D content creation, researchers from China have introduced DreamCraft 3D++. This new iteration significantly boosts both the speed and quality of generating complex 3D assets, reducing creation time to an impressive 10 minutes, a remarkable 20-fold increase in efficiency, and DreamCraft 3D++ maintains its predecessor’s multi-stage approach, but introduces key innovations that set it apart.

First, it replaces the traditional geometry sculpting process with a feed-forward multi-plane reconstruction model, achieving a staggering thousand times speed up. This innovation streamlines the workflow, eliminating the time-consuming optimization phase. Additionally, the framework features a novel, training-free IP adapter module for texture refinement. By dynamically selecting embeddings based on camera position, this module enhances both texture and geometry consistency, offering a fourfold increase in speed over the previous DreamBooth fine-tuning process. And experiments across diverse datasets demonstrate its superior ability to create intricate 3D assets, with realistic 360-degree textures, outperforming current state-of-the-art image to 3D techniques.

And the researchers have announced plans to open-source the entire DreamCraft 3D++ implementation, inviting the global community to explore and build upon their groundbreaking work. And finally, Adobe researchers just unveiled a new technology that creates realistic 3D images from just a few pictures, which they call the large-view synthesis model. This model uses advanced techniques to make this process faster and more accurate than ever before. LVSM comes in two versions. The first version turns input images into a compact format, which it then uses to generate new views. The second version skips this step and directly produces new images from the originals.

Unlike older methods that rely on complex 3D calculations, LVSM uses a more LVSM is quicker because of its efficient processing. While the second version delivers higher quality and can handle a wide range of situations without extra training, it even outperforms previous technologies by a significant margin. What’s truly impressive is that LVSM can create detailed 3D images using less computing power and can even work with just a single picture in many cases. This shows its ability to understand the depth and structure of a scene, making it a groundbreaking step forward in creating 3D visuals.

[tr:trw].