The role of Crypto Assets in the field of humanoid Bots

The data required for humanoid robots is cost-effective, scalable, and composable, while the token incentive model of Crypto Assets can fill the most urgent gaps currently.

Written by: @brezshares

Compiled by: AididiaoJP, Foresight News

Background Summary

General humanoid robots are rapidly moving from science fiction to commercial reality. Thanks to decreasing hardware costs, a surge in capital investment, and advancements in movement and flexibility, the AI computing field is brewing the next major transformation.

Although AI cloud computing and hardware facilities are becoming increasingly popular, providing a low-cost manufacturing environment for robotic engineering, the field is still limited by insufficient training data.

Reborn attempts to utilize DePAI for decentralized high-fidelity motion and synthetic data, and to build Bots foundational models. The project members come from the University of California, Berkeley, Cornell University, Harvard University, and Apple Inc.

Humanoid Robots: From Science Fiction to Reality

The commercialization of robots is not a new concept, such as the iRobot Roomba vacuum cleaner launched in 2002, or the popular Kasa pet cameras in recent years, but they typically have only a single functional design. With the development of artificial intelligence technology, robots are gradually evolving from single-function machines to multifunctional devices, capable of operating in unstructured environments.

In the next 5 to 15 years, humanoid robots will gradually expand from basic tasks such as cleaning and cooking to more complex fields such as concierge services, firefighting, and even surgical operations. Among the following three major trends, the vision is gradually becoming a reality:

• Rapid market expansion: More than 100 companies worldwide are engaged in humanoid robot research and development, including well-known enterprises such as Tesla, Unitree, Figure, Clone, Agile, etc.
• Breakthrough in hardware technology “Uncanny Valley”: The new generation of humanoid Bots moves smoothly and naturally, able to interact richly with humans. For example, the walking speed of Unitree H1 reaches 3.3 meters/second, far exceeding the human average of 1.4 meters/second.
• New Paradigm of Labor Costs: By 2032, the operating costs of humanoid robots are expected to fall below the wage levels of the average American workforce.

Bottleneck: Scarcity of Real-World Training Data

Although the field of humanoid Bots has broad prospects, its large-scale deployment will still be limited by the quality and scale of the training data.

Other AI fields (such as autonomous driving) have solved data issues through onboard cameras and sensors. For example, Tesla and Waymo train their autonomous driving systems using vast amounts of real driving data. Waymo is able to conduct real-time training for vehicles on the road and arranges for a Bots coach to be in the passenger seat during training.

However, consumers are less willing to actively provide data when using Bots, and they are unlikely to tolerate the existence of “robot nannies.” Therefore, humanoid robots must possess high performance at the time of factory release, making pre-deployment data collection a key challenge.

Although each training mode has its own scale unit, the scale of Bots training data differs from other AI domains by an order of magnitude:

• GPT-4: The training data includes over 150 trillion texts.
• Midjourney/Sora: Relies on billions of annotated video-text pairs.
• Bots dataset: The largest scale is only about 2.4 million motion segments.

This difference explains why bot technology has yet to establish a true foundational model, as data cannot be collected at all. Traditional data collection methods struggle to meet the demands:

1. Simulation Training: Low cost but lacks niche cases from the real world (i.e., “Sim2Real gap”).
2. Online Video: Lacks the force feedback or proprioceptive data required for Bots learning.
3. Real data collection: Requires manual remote control, the cost of a single machine exceeds 40,000 dollars and is difficult to scale.

Reborn attempts to acquire real-world data at low cost and high efficiency through a decentralized model, effectively addressing the Sim2Real gap issue.

Reborn: DePAI’s Full-Stack Solution

Reborn is committed to building a vertically integrated physical AI software and data platform, with the core goal of solving the data bottleneck of humanoid robots, but the vision goes far beyond that. Through proprietary hardware, multi-modal simulation infrastructure, and foundational model development, Reborn aims to become a full-stack driver in the field of intelligent humanoid robots.

ReboCap: Crowdsourced High-Fidelity Motion Data

ReboCap is a low-cost motion capture device developed by Reborn, with over 5,000 units sold and a monthly active user (MAU) count reaching 160,000.

Reborn achieves data collection with economic benefits superior to other alternatives.

Users generate high-fidelity motion data through AR/VR games and receive online incentives. This model not only attracts gamers but is also used by digital streamers to drive real-time digital avatars. This natural cycle of interaction completes scalable, low-cost, and high-fidelity data generation.

Roboverse: Unified Multimodal Simulation Platform

Roboverse is a multi-modal simulation platform designed to unify fragmented simulation environments. Current robot simulation tools (such as MuJoCo, NVIDIA Isaac Lab) have varying functions but are incompatible with each other, severely hindering research and development efficiency. Roboverse establishes standardized systems through simulators, creating a shared virtual infrastructure for the development and evaluation of robot models. By providing a unified development and evaluation platform, it enhances model compatibility.

Reborn Base Model (RFM)

Reborn Tech Stack

The most critical component of Reborn Full Stack is the Reborn Foundation Model (RFM). RFM is one of the first foundational models specifically designed for Bots, aimed at becoming the core infrastructure of DePAI. This is similar to traditional foundational models aimed at LLMs, such as OpenAI’s o4 or Meta’s Llama, but RFM is targeted at Bots.

ReboCap, Roboverse, and RFM have built a strong moat for Reborn. By combining the real data from ReboCap with the simulation capabilities of Roboverse, RFM is able to train high-performance models that adapt to complex scenarios, supporting diverse applications for industrial, consumer-grade, and research-type Bots.

Reborn is advancing the commercialization of technology and is currently conducting paid pilot projects in collaboration with Galbot and Noematrix, as well as establishing strategic partnerships with Yushu Technology, Booster Robotics, Swiss Mile, and Agile Robots. The humanoid robot market in China is growing rapidly, accounting for approximately 32.7% of the global market share. Notably, Yushu Technology holds over 60% of the global simulation robot market share and is one of the Chinese humanoid robot manufacturers planning to produce more than 1,000 units by 2025.

The Role of Crypto Assets in DePAI

Encryption technology is enabling DePAI to achieve a complete vertical stack.

Reborn is a leading project in the DePAI field

The DePAI project ensures open, composable, and permissionless scalability through token incentives, thereby achieving an efficient decentralized data collection and incentive model.

Reborn has not yet issued Tokens, but the token economics may accelerate the large-scale adoption of Reborn. Once the token incentive mechanism is launched, network participation is expected to grow rapidly:

1. Token Incentives: Users who purchase ReboCap can receive token rewards, while the Bots company pays for data, creating a positive cycle.
2. Edge Case Mining: Encourage users to contribute high-value edge case data through a dynamic incentive mechanism to bridge the Sim2Real gap.

Reborn’s DePAI Growth Flywheel

Data is the key

The true competitive advantage of humanoid robots lies in data and models. Specifically, it refers to the scale, quality, and diversity of the intelligent data used to train these machines.

The “ChatGPT Moment” of humanoid robots will not be led by hardware companies, as hardware deployment faces inherent challenges such as high costs and long cycles. The viral spread of robotics technology is essentially constrained by costs, hardware availability, and logistical complexities, whereas pure digital software like ChatGPT is not subject to such constraints.

Core Conclusion: Data is the Key to Victory

The real turning point will come from the data and model advantages after the cost reduction. The data required by humanoid robots is cost-effective, scalable, and composable, while the Token incentive model of Crypto Assets can fill the most urgent gap currently. Reborn turns ordinary people into “miners of motion data” through the Token incentive model of Crypto Assets.