Executive Summary
Humanoid general-purpose robots are rapidly transitioning from science fiction to reality. The continuous decline in hardware costs, ongoing growth in capital investment, and technological breakthroughs in motion flexibility and operational capabilities are three major factors that are continuously converging to actively promote the next significant platform iteration in the field of computing. Although computational power and hardware devices are becoming increasingly commoditized, providing cost advantages for robotic engineering, the industry is still constrained by the bottleneck of training data.
Reborn is one of the few projects utilizing decentralized physical artificial intelligence (DePAI) to crowdsource high-precision motion and synthetic data, and to construct foundational models for robots, placing it in a unique advantageous position to drive the deployment of humanoid robots. The project is led by a technically proficient founding team, whose members have academic research experience and faculty positions at institutions such as the University of California, Berkeley, Cornell University, Harvard University, and Apple Inc., reflecting both exceptional academic prowess and real-world engineering execution.
Humanoid Robots: From Science Fiction to Cutting-Edge Applications
The commercialization of robotic technology is not a new concept. Household robots, such as the iRobot Roomba vacuum cleaner launched in 2002, or the recently popular Kasa pet camera, are examples of single-function devices. With the development of artificial intelligence, robots are evolving from single-function machines to multifunctional forms, aimed at adapting to operations in open environments. In the next 5 to 15 years, humanoid robots are expected to gradually upgrade from basic tasks such as cleaning and cooking to eventually being capable of complex jobs like reception services, firefighting, and even surgical operations. Recent developments are turning humanoid robots from fiction into reality.
Market Dynamics: Over 100 Companies Engaged in Humanoid Robotics (e.g., Tesla, Utree Technology, Figure AI, Clone, Agile, etc.)
Hardware technology has successfully crossed the uncanny valley: the new generation of humanoid robots demonstrates natural and fluid movements, allowing them to achieve human-like interactions in real-world environments. For example, Utree’s H1 can walk at a speed of 3.3 meters per second, far exceeding the average human walking speed of 1.4 meters per second.
(Note: The Uncanny Valley theory is a psychological theory describing human emotional responses to non-human entities, such as robots, dolls, and virtual images.)
The cost paradigm for humanoid robots is expected to drop below the U.S. labor wage level by 2032.
Development Bottleneck: Training Data in the Real World
Despite clear advantages in the humanoid robotics field, issues of low-quality and scarcity of data continue to hinder large-scale deployment. Other artificial intelligence technologies, such as autonomous driving, have fundamentally addressed data issues through cameras and sensors mounted on existing vehicles. For instance, fleets of autonomous driving systems like Tesla and Waymo can generate billions of miles of real-world driving data. During development, Waymo had real-time trainers in the passenger seat to provide immediate training.
However, consumers are unlikely to accept the existence of “robot babysitters.” Robots must possess high performance right out of the box, making data collection prior to deployment critical. All training must be completed before commercial production, and the scale and quality of data remain ongoing challenges. Although each training mode has its own scaling units (e.g., tokens for large language models, video-text pairs for image generators, and motion clips for robotics), the comparison below clearly reveals the magnitude of the gap in data availability faced by robotic technology:
- The training data scale for GPT-4 exceeds 150 trillion text tokens.
- Midjourney and Sora utilize billions of labeled video-text pairs.
- In contrast, the largest robotic dataset contains only about 2.4 million interaction records.
This disparity explains why robotic technology has not yet achieved truly foundational models like large language models; the key lies in the incomplete data foundation. Traditional data collection methods struggle to meet the scaling needs of training data for humanoid robots. Existing methods include:
- Simulation: Low cost but lacks real boundary scenarios (the gap between simulation and reality).
- Online Videos: Cannot provide the necessary proprioceptive and force feedback environments for robot learning.
- Real-world Data: While accurate, it requires remote control and human closed-loop operations, leading to high costs (over $40,000 per robot) and a lack of scalability.
Training models in virtual environments is cost-effective and highly scalable, but these models often struggle during real-world deployment. This issue is referred to as the sim-to-real gap (Sim2Real). For example, a robot trained in a simulated environment may easily grasp perfectly illuminated and smooth-surfaced objects, but it often falters when facing cluttered environments, uneven textures, or various unexpected situations that humans commonly encounter in the real world.
Reborn provides an economical and efficient way to crowdsource real-world data, enhancing robot training and addressing the “sim-to-real gap” challenge.
Reborn: A Full-Stack Vision of Decentralized Physical AI
Reborn is building a vertically integrated software and data platform aimed at applications of embodied intelligent robots. The company’s core goal is to solve the data bottleneck issues in the humanoid robotics field, but its vision extends far beyond that. Through the combination of self-developed hardware, multimodal simulation infrastructure, and foundational models, Reborn aims to become a full-stack driver of embodied intelligence.
The Reborn platform starts with its proprietary consumer-grade motion capture device, “ReboCap,” establishing a rapidly expanding ecosystem of augmented reality and virtual reality gaming. Users are incentivized to provide high-fidelity motion data in exchange for online rewards, driving the continuous growth of the platform. Currently, Reborn has sold over 5,000 sets of ReboCap devices, with monthly active users reaching 160,000, and has established a clear growth path to surpass 2 million users by the end of the year.
Notably, this growth has entirely stemmed from organic development: users are attracted by the entertainment value of the games themselves, while streamers leverage ReboCap for real-time motion capture of their digital avatars. This spontaneously formed virtuous cycle has enabled scalable, low-cost, high-fidelity data production, making Reborn’s dataset a sought-after training resource among leading robotics companies. The second layer of the ReBorn software stack is Roboverse: a multimodal data platform unifying fragmented simulation environments. Currently, the simulation field is highly fragmented, with tools like Mujoco and NVIDIA Isaac Lab operating independently. While each has its advantages, they cannot interoperate. This division delays the development process and exacerbates the gap between simulation and reality. Roboverse creates shared virtual infrastructure for developing and evaluating robotic models by standardizing multiple simulators. This integration supports consistent benchmark testing, significantly enhancing the system’s scalability and generalization capabilities.
Roboverse enables seamless collaboration. The former extensively collects real-world data while the latter builds simulated environments to drive model training, showcasing the true power of Reborn’s distributed physical intelligence network. The platform is creating a developer ecosystem for physical artificial intelligence that goes beyond mere data acquisition, extending its functionalities into actual model deployment and commercial licensing.
Reborn Foundational Model
The most critical component of the Reborn technology stack may be the Reborn Foundational Model (RFM). As one of the first foundational models for robotics, this model is being developed as the core system of emerging physical artificial intelligence infrastructure. Its positioning is akin to traditional large language foundational models, such as OpenAI’s GPT-4 or Meta’s Llama, but targeted towards the robotics field.
The three core components of the Reborn technology stack (ReboCap data platform, Roboverse simulation system, and RFM model licensing mechanism) collectively build a solid vertical integration moat. By combining crowdsourced motion data with a powerful simulation system and model licensing framework, Reborn can train a foundational model with cross-scenario generalization capabilities. This model can support a diverse range of robotic applications in industrial, consumer, and research fields, achieving generalized deployment under massive and diverse data.
Reborn is actively advancing the commercialization of its technology, launching paid pilot projects with Galbot and Noematrix, and establishing strategic partnerships with Unitree, Booster Robotics, Swiss Mile, and Agile Robots. The humanoid robotics market in China is experiencing rapid growth, accounting for approximately 32.7% of the global market. Notably, Utree Technology holds over 60% of the global quadruped robot market and is one of six Chinese manufacturers planning to produce over 1,000 humanoid robots by 2025.
The Role of Cryptocurrency Technology in the Physical AI Technology Stack
Cryptographic technology is building a complete vertical stack for physical world artificial intelligence.
Although these projects belong to different layers of the physical AI stack, they share a commonality: they are all 100% DePAI projects. DePAI integrates token incentives throughout the technology stack, creating an open, composable, and permissionless scaling mechanism, which is precisely the innovation that makes the decentralized development of physical AI a reality. Reborn has yet to issue tokens, making its organic growth even more remarkable. Once the token incentive mechanism officially launches, network participation will serve as a critical link to accelerate the DePAI flywheel effect: users purchasing Reborn hardware (ReboCap capture devices) will receive incentives from the project team, while robotics R&D companies will pay contribution rewards to ReboCap holders. This dual incentive will drive more people to purchase and utilize ReboCap devices. At the same time, the project team will dynamically incentivize the collection of high-value customized behavior data, thereby more effectively bridging the technical gap between simulation and real-world applications (Sim2Real).
The “ChatGPT moment” in the robotics field will not be triggered by robotics companies themselves, as hardware deployment is far more complex than software. The explosive growth of robotic technology is inherently limited by cost, hardware availability, and deployment complexity—barriers that do not exist in purely digital software like ChatGPT. The turning point for humanoid robots lies not in how impressive prototypes are, but in reducing costs to an affordable range for the masses, similar to the proliferation of smartphones or computers in the past. As costs decline, hardware will become the ticket to entry, with the real competitive advantage lying in the scale, quality, and diversity of the data and models used to train the machines.
Conclusion
The revolution of robotic platforms is unstoppable, but like all platforms, its scalable development relies on data support. As a high-leverage bet, Reborn firmly believes that cryptographic technology can fill the most critical gap in the AI robotics technology stack: its robotic data solution, DePAI, possesses cost-effectiveness, high scalability, and modular characteristics. As robotic technology becomes the next frontier for AI, Reborn is transforming the general public into “miners” of motion data. Just as large language models require text tokens for support, humanoid robots need massive sequences of motion for training. Through Reborn, we will break through the final bottleneck, realizing the leap of humanoid robots from science fiction to reality.
This article is collaboratively republished from: PANews