Humanity's fascination with automata predates myths and bleed into history. From Hephaestus’s golden handmaidens, who moved with living grace and voice to assist the limping god at his forge, to the Han-dynasty musical puppets troupes, the urge to animate matter has always mirrored our desire to externalize thought. Such machines served as more than entertainment, but also acted as extensions of divine and imperial power.

The commoditization of steam power and the advent of industrial revolution realized machines mass in gross forms. But for all these engines and gears, it birthed only the body in a strict routine, but not the mind.

The innovations in the last year however, shows promise of machines capable of generalization. Now as AI permeates the domain of robotics, we approach convergence, as we try to grant reasoning to embodiment. Humanity's stands at the precipice of our oldest dreams, machines that can not only do, but also think.

Large Language Models (aka ChatGPT) has reshaped the white-collar world within a few years. Paradoxically, humanity found it easier to automate average reasoning tasks over physical work. Physical labor has remained a stubborn frontier, limited by locomotion and manipulation issues.

Pioneering research of late however, has shown that foundational robotic models, trained on heterogeneous datasets, can generalize across tasks, environments, and embodiments to a certain extent. A single model capable of performing a wide range of physical tasks is no longer theoretical. With sufficient data scale, robots can even potentially exhibit a kind of fluid intelligence, allowing them to grasp, push, and navigate by understanding space.

The internet has made the world smaller, compressing billions of interactions into code, unanchored from matter atomically. Progress drifted skyward, abstracted into the cloud. Embodied AI reintroduces gravity into the digital loop. It drags intelligence back into the tangible spatially, demanding for models to exist and navigate the world materially. It is the return of flesh to code, closing the mechanical - digital loop that was kickstarted during the industrial revolution.

On the advent of a robotics foundational model, we believe the time is right for artificial intelligence to extend into physicality, as we are on the cusp of a artificial digital and physical revolution;

1. Blockchain: solves trust at scale, allowing for massive coordination immutably without intermediaries as a fair backend

2. AI: solves reasoning, allowing for problem solving of the abstract with minimal (or no) supervision

3. Robotics: now approaches physical execution with cheaper costs, allowing for perpetual labor without fatigue and precise performance

The combination of which, forms a closed circuit of autonomy, where transactions, thoughts and actions can propagate, a self-driving system producing value we term as aGDP (agentic GDP).

aGDP: the aggregate output of humans, agents, and machines cooperating across both digital and physical domains

Each new frontier of autonomy expands this economic surface area, from trading to entertainment and beyond. Yet robotics doesn't merely add another frontier, it opens an entirely new dimension, physicality. Digital agents (white-collar) and embodied agents (blue-collar) will form the full spectrum of humanity's productivity, and robotics is the lever by which digital productivity will translate into material aGDP.

Our journey into robotics began through our venture arm (purely out of our balance sheet to prevent conflict of interests) a year ago, investing into a range of frontier robotics startups. Those early bets have materially shaped our understanding of how fragmented and capital-intensive the field is. Through our capital formation launchpad Unicorn, we have begun to solve this bottleneck. By aligning incentives and liquidity, we enabled builders to fund ambitious ideas that traditional markets often overlook. Virtuals is now positioned as the pre-eminent AI capital formation platform, one that can collectively finance robotic deployments and meaningfully compress onchain WACC (weighted average cost of capital) by leveraging speculation premia.

Yet financing alone is not enough. To truly accelerate embodiment, we must also complete the loop, specifically in data. Training policies and models require more than scale, but also spatial depth, specifically how objects move and interact in the 3-dimensional space. Today, such data remains scarce, and without massive, diverse, real-world datasets, embodied AI will remain in limbo. To teach robots how to move, we must first let them see, which is what we have been building for the better part of this year - SeeSaw

SeeSaw: humans see (and recorded), robots saw (and learned)

SeeSaw is a mobile iOS app egocentric video data collection app by crowd-sourcing task oriented videos of hand-object interactions, by gamifying the process. Users complete quests by recording them in-app, and get rewarded with rewards and points. On the backend, we have an extensive infrastructure to ensure that data collected is of par, supported by our tech partner, BitRobot:

1. Egocentric data with spatial depth: leveraging the iPhone's LiDAR and IMU sensors, which is crucial for accurate scene reconstruction

2. Automated VLM pipeline: automatically scores and quantifies the utility of each video for training policies

Combined with our expertise in structuring incentives systems, we believe SeeSaw to be a viable solution in scaling a diverse real-world dataset in millions of homes and work environments. Through this vector, Virtuals will be the perceptual substrate for which embodied intelligence learns.

A new world - where agents drive material GDP - now lies within reach. By aggregating coordination, reasoning, and execution, the embodied network can translate digital intelligence into physical productivity. The future ascent of humanity towards the stars will not be led by humanity alone, but also with our agents, autonomous entities that can think, do, and transact. Many other research and engineering bottlenecks remain in robotics, but crypto can materially accelerate 2 core pillars; capital formation and data. Together, they help close the loop to bring intelligence into physicality.

Just as Nāgārjuna sought enlightenment through the middle Way, avoiding both eternalism and nihilism, we too take the middle road for robotics in crypto

In pursuit of this future, we have chosen the middle Way (in crypto) for robotics. Rather than building frontier models or hardware ourselves, we focus on the invisible levers that make embodiment inevitable. By constructing the data and capital foundations of the ecosystem, we accelerate progress at the outskirts, building the layer by which intelligence can accelerate and perform.

This philosophy manifests through the tools we create, systems that empowers. Through Unicorn, we are reimagining how capital flows into frontier technologies. Via SeeSaw, we are redefining how the world is captured and learned. As these vectors intertwine, Virtuals evolves beyond an onchain AI agent platform into a fullstack engine for all artificial intelligence, embodied or otherwise.

These are but small steps to a larger continuum, but if played well, can accelerate the deployment of artificial reasoning into the real world.

If the last decades was defined by information, the next will be by embodiment, the moment when thought reclaims form.