Code Moment | Independent Variable Robotics Closes Four Consecutive Funding Rounds, Valuation Exceeds 20 Billion Yuan

Recently, MaHui portfolio company and Embodied Artificial Intelligence firm Independent Variable Robotics (X Square Robot) completed consecutive B+, B++, and C funding rounds, with investors including China Mobile's "Chain Leader" Fund, HSG, IDG Capital, Source Code Capital, Fortune Venture Capital, CICC Capital, China Insurance Investment, and more than 30 top-tier institutions across various sectors.

This follows the company's Series B round completed in late April, marking another series of major financings closed within just over two months, with its overall valuation surpassing RMB 20 billion. The proceeds will fund R&D on foundational embodied intelligence models, large-scale data collection, and continuous upgrades to robot hardware.

Source Code Capital participated in Independent Variable Robotics' Series C round, formally becoming a shareholder. Embodied Artificial Intelligence has long been a core focus area for Source Code Capital, with investments spanning today's hottest categories: humanoid robots, consumer service robots, and industrial robots, as well as general-purpose embodied foundation models. Its portfolio already includes Unitree, Galaxy Universal, Booster Robotics, Woan Robotics (06600.HK), HAI ROBOTICS, Mech-Mind, Standard Robots, Cloud Whale, Keenon, Starmax Innovation, Oak Robotics, and others. Investing in Independent Variable Robotics represents another critical piece of Source Code Capital's embodied intelligence ecosystem.

1

Full-Stack In-House Development: The Trinity of Model, Data, and Infra

In April 2026, Independent Variable Robotics released WALL-B, the world's first embodied foundation model built on a "World Unified Model (WUM)" architecture. Unlike VLA models that stitch together separate vision, language, and action modules, WALL-B puts vision, language, action, and physical prediction capabilities into the same network and trains them jointly from scratch as an integrated whole — eliminating boundaries between modules and the data transfer overhead between them. This gives WALL-B three key advantages: native multimodality and stronger spatial reasoning; the ability to understand physical laws and predict environmental evolution; and a memory capacity that enables self-improvement through environmental interaction. This is a ground-up rewrite of both architecture and training paradigm, and a landmark moment in embodied foundation models' evolution toward native multimodality.

Beyond this, over the past three months Independent Variable Robotics has demonstrated its absolute technical strength through a concentrated series of open-source releases and product launches:

Its open-source WALL-OSS-0.5 unifies vision and action representation, with action tokens corresponding to observable physical changes. According to the company, even pre-training alone yields task capabilities comparable to post-training results, outperforming mainstream open-source models such as Pi 0.5 on multiple real-robot tasks. Its recently released world model WALL-WM aligns multimodal data — language, vision, and action — by "events" rather than uniform time sampling. Published benchmarks show it scoring significantly above mainstream models including Pi 0.5 and DreamZero on foundational tasks, reasoning tasks, dexterous manipulation, and generalization scenarios.

Beyond models, Independent Variable Robotics has built out the two foundational pillars of data and infrastructure. On the data side, the company developed an industrial-grade data pipeline covering large-scale collection, cleaning, auto-labeling, quality control, and augmentation, alongside proprietary data synthesis models that replicate rare, complex, or dangerous real-world scenarios. Its companion XR Zero series of hardware-free data collection solutions supports full-modality data replay and transferability; according to the company, this has been validated in closed model loops and can reduce training data costs by roughly 95%.

On the infrastructure side, Independent Variable Robotics developed its own distributed training and high-performance inference framework purpose-built for embodied foundation models, constructing a closed data feedback loop of "robot execution → data collection → retraining → deployment" with an online reinforcement learning system that lets models continuously absorb new data and self-evolve through real-world deployment.

2

Dual-Track Commercialization: From Homes to Production Lines, Toward General Embodied Intelligence

On the hardware side, Independent Variable Robotics has self-developed and mass-produced two high-performance robots, Quantum One and Quantum Two, achieving full in-house design of core components including robotic arms, joint modules, motor drivers, and main controllers, with deep algorithmic optimization — providing the physical platform for robots to enter both homes and industrial production lines.

In the consumer market, Independent Variable Robotics has launched an ambitious program to bring robots into households at scale. Starting in March, the company partnered with 58 Daojia to offer intelligent cleaning services, with robots collaborating with human cleaners to complete household tasks — a landmark event in embodied intelligence technology's move toward larger-scale commercial application, with the potential to transform how families live in the future.

Beginning in May, Independent Variable Robotics went further with its "X Family Member" program, offering users robot residency services lasting up to one month to meet more diverse and spontaneous household needs — receiving widespread positive feedback from users. This marks the first time embodied robots have stepped off the "demo stage" and entered homes at scale to face real user demands.

On the enterprise side, Independent Variable Robotics is collaborating with auto parts manufacturers, logistics companies, and others, with its embodied foundation models and robots already deployed in validation scenarios including precision operations on industrial manufacturing lines and high-requirement tasks such as logistics sorting and package feeding.

Independent Variable Robotics' vision is "to let intelligent robots serve humanity every day." To this end, the company has built a complete closed loop from algorithms to full robots through self-developed world unified models, embodied robot hardware, data collection solutions, and data pipelines. In the embodied intelligence industry, robot hardware technology is gradually converging; only the "brain" can give robots generalization capabilities and make them truly usable productive forces in homes and factories. As its name suggests, Independent Variable Robotics aims to be the "independent variable" that actively drives industry transformation, pioneering the future of "general embodied intelligence" through first principles.