GitHub

He's the one who wrote it.

I'll be the editor of Ziss.

The kryptons were informed that the Astrobot company has recently completed a B-wheel series of financing, with three rounds of cumulative financing exceeding $1 billion over three months, with investments from older shareholders, such as Lianghukko-Industries Second Generation Fund (Bohua Capital Management), Yangzhou Dragon cores, Chinese cumulation, Middle Cobreda, Kodeh Education, a head-listed enterprise and State investment.

The current valuation of stardust intelligence has exceeded a billion dollars, and this is another ten billion unicorn born in Shenzhen. Prior to this, the company's investment portfolio had seen the appearance of telecommunication, Ali and byte institutions.

In terms of commercialization, Stardust intelligence has simultaneously placed several industrial scenarios on cooperative orders, including the order for thousands of industrial and commercial services in Central Cote d ' Ivoire and the promotion of sea access, as well as the construction of a billion-dollar centre for applied innovation with Jiangdu through the open area of the river.

A fundamental shift in the logic of smart track competition is taking place. Industry farewells focus on the rough contrasts of stage pre-set demonstrations, turning to the complex problem of landings in real physical environments. The technical route has not yet shrunk, and different dimensions of moving programmes, model structures, data strategies, etc. have been explored and trade-offs among manufacturers based on different understandings of landing paths.

Founded in 2022, the founder and CEO Leager has more than 17 years of research and development and design experience in the field of AI and robots, has led the development and development of a plethora of new types of robots, and has served as the head of a team of staff and architects of Unit 1 of the Tunte Robotic Laboratory, and 100-degree robots.

In the early stages of entrepreneurship, Jay clearly proposed the “Design for AI” path, which would define the body of a robot that fits the learning and evolution patterns of the large AI model, and then the AI algorithms and operating systems that match that body, so that robots can “think like people and work like people”. On the basis of this idea, the stardust intelligence has built a whole-of-house self-study system for the “AI model - with a OS-lined home” system.

Stardust smart T1 new product (graph source/enterprise)

How can this robot think and act like a human being? In this regard, the stardust intelligence has adopted a self-research matrix model that relies on data efficiency rather than data size-driven capacity enhancement.

Real physics has a long and easily neglected characteristic: not all actions require in-depth reasoning. For example, in a home setting, with microwaves, toys and delivery tools, these high-frequency operations rely more on rapid ingenuity responses, while long-range tasks such as “collating and cooking kitchens” require modeling for steps to dismantle, understand the environment and process anomalies. Different needs correspond to the timescale and calculator consumption required, large models are capable of reasoning but slow to respond, end-to-end action models are responsive but lack planning capacity.

Stardust intelligence is thinking that instead of striking a balance on a model, it is better to let the model do its job. At the bottom end of the model, the stardust intelligence presents the end-to-end whole-of-body VLA base model Lumo, which carries on the need for generic reasoning for higher dimensions.

As a broad-based global model, Lumo addresses tasks such as complex semantic understanding, abstract command dismantling and the generalization of unknown scenarios. Model training uses "pre-training + real-time alignment" to build a basic cognitive system by learning generic mission logic and step-dismantling through big data, and then fine-tuned real-format data collected on a tether-dependant multi-mode data to enable the flow of action from model "thinking" to robotic "practical".

Lumo-1 for long-sequence complex tasks (source/enterprise)

Based on this training logic, Lumo can demonstrate a greater generalization of the field in unknown objects, unfamiliar environments, vague abstract instructions, etc. Hard krypton understands that the future version of Lumo will also be integrated into the world model prediction capabilities and will further enhance robotic prejudication and reasoning efficiency in the dynamic environment.

At the end of the commercial framework, based on the dual-system-of-human cognitive theory, the stardust intelligence developed the first DuoCore fast and slow synergetic framework for full-body mobility, which is almost consistent with the Helix structure subsequently released by the United States company Figure.

The architecture divides robotic intelligence into two separate and synergistic capability systems. The fast-system master plays the instinctive real-time response, is responsible for the dynamic basic actions of millisecond-grade attitude fine-tuning, on-site shielding, joint flexible buffers, which match the environmental disturbances that occur at any time in the real scene; the slow-system master takes the cognitive depth planning, focuses on long-time sequenced missions, cross-spatial path planning and global strategy generation, and supports the whole body-to-body work of bending, crouching, moving and fine working with both arms.

DuoCore retail application of SSI commercial model architecture has been scaled down (source/enterprise) in six cities across the country

The two systems, which operate in a single OS, are accompanied by slow-system output of the overall mission programme and trajectories, enabling the fast system to modify aerophysically appropriate and non-standard environmental changes in real time; if there is a sudden-onset risk, the fast system can automatically trigger an emergency avoidance and, in conjunction with slow-system updates of mission planning, achieve dynamic error correction.

Distinguished from the traditional AI “Max Feeding from Zero Learning” model, the DuoCore framework builds on the logic of human skills migration, replicating existing experience in cross-scene missions and addressing the downside challenge of real-time, dynamic, and efficient learning of robotic scenarios.

Stardust intelligence is the world ' s first enterprise to achieve a rope-driven AI robotic mass production. It mimics the motor logic of human muscles, stretches through the back of an electric machine, and stretches through a tether tow to control the joints, creating a “frequent coupling” character that preserves sufficient operational rigour and has a soft buffer and shock absorption capacity.

The higher payload capacity, lower backlash and inertia, and more compact airframes than traditional rigid cortex structures, allow robots to ensure the safety of close interaction while maintaining high-dynamic operations and high-profile performance, and broaden the application imagination from commercial to household.

Comparison of different drivers (source/enterprise)

When robots carry out their mission, rope drives lower frictions and more continuous motion, which is the key data on which the AI megamodels rely for learning about real physical interaction patterns, as compared to the reduced dentures and mechanical noises of traditional rigid institutions, which are less costly and more complete in transmitting quality real-time control data to AI without loss.

Entering the productization phase, based on a modular design of a unique tether structure of stardust, robotics are broken down into separate modules, such as arms, torso, legs, etc., and any component is damaged, which can be quickly recovered by replacing the corresponding module without the need for a complete return to the plant, which reduces the customer ' s downtime cost and use threshold. Its tethers had already started a 1,000-stage delivery by the end of 2025.

Stardust smart T1 new, at a start-up price of $899 million, to drive deployment to scale (source/enterprise)

Stardust intelligence is synchronizing the multi-formula product to land. In recent days, the company has issued T-series T1, at a start-up price of $899 million, which allows for continuous fine work, such as steak frying, laundry, business dressing, chemical experiments, car selection, car charging, etc.

At present, stardust intelligence has developed a mass delivery capacity in areas such as scientific research, commercial services, recreational performances and industry.