Independent Variable Robotics Closes Four Rounds in Two Months, Valuation Tops 20 Billion Yuan | Unity Ventures Portfolio News

Unity Ventures early-stage portfolio company Independent Variable Robotics has announced four consecutive funding rounds, pushing its post-money valuation past 20 billion RMB with all capital fully closed and received.

Among domestic embodied AI companies, Independent Variable is the only one known to have been lead-invested by all four major internet giants, each continuing to double down. Meituan led the Series A, Alibaba led the A+, ByteDance led the A++, and Xiaomi led the Series B.

This year, the company has unveiled a string of firsts: WALL-B, the world's first embodied large model built on a World Unified Model architecture; WALL-OSS-0.5, an open-source embodied foundation model; and WALL-WM, the world's first world model with event-level prediction capability. It has also built a data factory and developed its own robot hardware, holding all three pieces — model, data, and body — to form a self-improving loop.

Just two months after announcing its Series B led by Xiaomi at a late-April launch event, Independent Variable closed three more rounds — B+, B++, and C — pushing its valuation past 20 billion RMB with all capital fully closed and received.

In the embodied AI race, funding announcements alone are no longer news. What truly stands out is how quickly and decisively the money hit the bank.

Four rounds in just over two months, all closed — even in today's most funding-frenzied sector, that's rare.

Even more striking is the investor lineup.

More than 30 confirmed institutions span four camps: internet giants, corporate strategics, state-backed funds, and top-tier VC. China Mobile, China Insurance Investment, HSG, IDG Capital, Source Code Capital, Fortune Venture Capital, and CICC Capital are all on the cap table, with Xiaomi investing across three consecutive rounds.

Thirty-plus institutions, tens of billions in capital, all betting on the same thing: the embodied brain.

The Big Four Internet Giants Go All In

If there's one signal that matters most in this funding wave, it's the collective conviction of the internet giants.

Among domestic embodied AI companies, Independent Variable is the only known case where all four major internet firms have each taken the lead in a round and continued to add.

Meituan led the A, Alibaba the A+, ByteDance the A++, and Xiaomi the B — a "grand slam."

Xiaomi, in particular, has kept doubling down across the B, B+, and B++ rounds with barely a pause.

Corporate strategics are piling in too.

Across the latest four rounds, 58 Group, Shenyang Auto (Shenyang Automotive Industry Investment Fund), Chery Automobile (Guohai Innovation Capital), and HONOR (Shenzhen AI Terminal Fund) have all come aboard, covering household services, auto manufacturing, and consumer electronics.

These aren't bets on concepts — they're bets on robots that will actually walk into their business operations.

58 Group maps to household services, Chery and Shenyang Auto to automotive manufacturing, HONOR to consumer electronics. What they want to know is whether embodied AI can solve real problems in real scenarios.

Already, Independent Variable has entered 58 Home's service operations and a German luxury automaker's parts production line, moving from lab to home services and industrial manufacturing.

State-backed capital is also stepping up.

Beyond new investors like SDIC Innovation, China Insurance Investment, Jiangsu High Investment, Shenzhen Investment Holdings Capital, and Bao'an District Guidance Fund, returning investors Guokai Technology Venture Investment and CASID have re-upped, with China Mobile adding across two consecutive rounds.

The message isn't hard to read.

With embodied AI written into the "15th Five-Year Plan" as a core future industry, state assets, local funds, and SOE capital are accelerating toward the sector's leading companies.

These investments bring more than money — they signal long-term support in local industrial resources and supply chain coordination.

Market-driven VC hasn't sat this out either.

HSG, IDG Capital, Fortune Venture Capital, CICC Capital, Source Code Capital, and Yida Capital have all placed bets. HSG, notably, has followed from the A+ round last September through today, rarely missing a round.

Counting earlier financing, Independent Variable has completed more than ten rounds in two and a half years since founding.

Internet giants, corporate strategics, state-backed funds, and top VC — all four camps continuously heavy-weighting the same company. You won't find this capital structure elsewhere in China's embodied AI track.

So what are all these discerning investors actually betting on?

Betting on the Brain

They're betting on the "brain" of embodied AI.

Independent Variable founder Qian Wang made an early call: embodied intelligence models are not an extension of language models, but a parallel class of foundation models in their own right.

The reason is straightforward. For robots to truly enter the physical world, executing commands isn't enough. They must understand environments, predict changes, and know what might happen next.

Over the past year, this has become near-consensus across the industry, and the capability enabling it is the world model.

Simply put, a world model is an AI specialized in predicting "what happens next."

If large language models predict the next token, world models predict the next frame — and the physical changes within it. For robots, it's a brain that can mentally simulate the future.

In April, Independent Variable unveiled WALL-B, the world's first embodied large model built on a World Unified Model (WUM) architecture.

Unlike traditional approaches that split perception, decision-making, and action into separate modules串联起来, WALL-B unifies vision, language, action, and physical prediction within a single network, trained jointly from scratch.

Information no longer passes through layers of handoffs; the model learns the relationships between capabilities directly.

This gives WALL-B three core abilities: native multimodal understanding, physical world prediction, and continuous learning through environmental interaction.

Yet more attention-grabbing than the model itself was the goal Independent Variable set for it.

At the launch, the company announced its intention for WALL-B-powered robots to live long-term in real homes.

Compared to a lab, an ordinary household is genuinely complex: objects move daily, family behavior is unpredictable, almost no two days are alike.

Whether it can operate stably in such an environment long-term has become the most direct — and most demanding — test of WALL-B's generalization.

The Embodied Brain's Rapid Iteration

The willingness of capital to keep coming back round after round also owes to Independent Variable's relentless pace of technical progress.

In just the past month-plus, the company has released two core models.

First: WALL-OSS-0.5, an open-source embodied foundation model. It had only completed pre-training with no task-specific post-training, yet achieved results close to many competitors' finished models.

Across 17 real-world robot tasks, four achieved over 80% autonomous completion rates, surpassing mainstream models including Pi 0.5 open-sourced by overseas star company Physical Intelligence on both manipulation and reasoning.

Second: WALL-WM, the world's first world model with event-level prediction capability.

Where most world models sample uniformly in time, slicing video frame by frame, WALL-WM takes a different approach: it understands the world by events, aligning language, vision, and action around the same occurrence. This lets the model more easily learn true causal relationships between modalities, enabling more accurate predictions of what happens next in the physical world.

Beyond models, what's truly scarce in embodied AI is data.

To address this, Independent Variable built its own data factory, developing a full data production pipeline from collection through cleaning, annotation, and quality control — all automated and scaled.

Using its self-developed data collection device XR Zero G0, the company has cut embodied training data collection costs by 95%.

Going deeper still: robot hardware itself.

The two generations of robots, Quantum One and Quantum Two, provide platforms for continuous model iteration and real-world validation.

With this, model, data, and body — all three pieces rest in Independent Variable's own hands, forming a self-improving loop.

Two People Who Bet Early on Independent Models

Going back to the beginning, Independent Variable placed its bet on a path different from most from day one.

Founder and CEO Qian Wang holds bachelor's and master's degrees from Tsinghua University and a PhD from the University of Southern California, where he researched Robotics Learning and was among the earlier researchers to introduce attention mechanisms into neural network architectures.

Co-founder and CTO Allan Wang holds a PhD in computational physics from Peking University and previously led the large model team at IDEA Research Institute, spearheading the release of multiple open-source large models.

The "independent foundation model" thesis — Wang has held it unwaveringly since the company's founding.

"Independent" doesn't mean different application scenarios. It means confronting an entirely different problem.

The continuous states, causal relationships, and action feedback of the real physical world. Its modeling objectives, training methods, and evaluation systems cannot be borrowed wholesale from language models serving the virtual world.

Based on this conviction, Independent Variable has consistently pursued self-developed fully end-to-end general embodied large models, tackling model, data, and robot body together.

Because when general embodied intelligence reaches its endgame, the contest isn't just model versus model, or body versus body — it's whether model, data, and hardware can form a continuously iterating loop. Any weak link becomes the ultimate ceiling.

This means that rather than standing on existing infrastructure, Independent Variable chose a heavier, slower, harder path: starting from the bottom and keeping every building block in its own hands.

In the past, embodied AI funding favored spreading bets across various robot bodies. Now, increasingly, capital is concentrating on the "brain."

Thirty-plus institutions crowding into one company aren't betting on yet another robot — they're betting on whether the world model in its head can truly walk into factories and into millions of homes.

That answer still needs time. But the capital willing to bet early has already written its answer: 20 billion RMB.

Source: QbitAI

「Aether AI」完成2000万美元融资，以因果世界模型驱动下一代 AI 范式变革 | 九合系新成员

「基点起源」完成数亿元融资，以"全要素大模型"构筑工业AI新基座 | 九合系融资新闻

「欧拉万象」完成数亿元天使轮融资，以 "可养成" 范式重构家庭具身智能的物理世界数据闭环 | 九合系融资新闻