"RoboParty" has completed nearly 500 million yuan in Series Angel++ and Series Pre-A financing, building an open-source full-stack embodied intelligence platform

Recently, the open-source full-stack embodied intelligence platform "RoboParty" has consecutively closed nearly 500 million RMB in Series Angel++ and Pre-A financing rounds. The Series Angel++ round was led by continued investments from existing shareholders including Shunwei Capital, Matrix Partners China, Xiaomi Strategic Investment, and SenseTime Guoxiang, while the Pre-A round was exclusively funded by CATL. Lightsource Capital served as the incubator and exclusive financial advisor.

The repeated continuous investments from long-term shareholders validate RoboParty's rare technical strength, exceptional product taste and iteration speed, as well as its robust industry ecosystem positioning and capital momentum. CATL's strategic backing further expands RoboParty's industrial imagination space. Following Xiaomi Strategic Investment, CATL Strategic Investment's major re-investment in RoboParty signifies that RoboParty is no longer defined solely by open-source attributes — it has evolved into an embodied intelligence platform with the potential to penetrate real manufacturing, energy, supply chain, and complex operational scenarios. CATL and the company will jointly explore all possible forms of collaboration across the entire industrial value chain.

Open Body × Party OS × Developer Ecosystem

Over the past few years, the embodied intelligence track has experienced rapid growth. As supply chains mature, hardware form factors converge, and leading players accelerate mass production, the market window for emerging startups is narrowing rapidly. An increasing number of companies enter the sector via world models or embodied foundational models, yet many "brain-focused" firms lack proprietary robot hardware, leaving their training, validation, and deployment workflows dependent on third-party hardware platforms. This inherently limits their capabilities in data closed-loop operations, real-machine feedback, model iteration, and real-world scenario implementation. Companies that integrate full in-house R&D capabilities, vertical integration and mass production expertise, end-to-end software-hardware full-stack proficiency, and open ecosystem building skills are extremely rare — RoboParty stands out as one of the very few remaining startups positioned to secure a seat at the industry's core table at this critical juncture.

What sets RoboParty apart is its identity as an open-source full-stack embodied intelligence platform with rare proprietary hardware capabilities. Founded just over a year ago, the company has completed foundational development across joint modules, full robot platforms, motion control systems, SDKs, and open-source development frameworks, and is now entering the mass production ramp-up phase. This in-house hardware capability forms a solid commercial safety buffer for RoboParty, laying the groundwork for it to operate in the real world, support a thriving developer ecosystem, and accumulate valuable robotics datasets. Furthermore, the company has made strategic investments in BFM Off-policy Trackers and Humanoid Foundation Models (HFM), and in January 2026, recruited multiple top talents from Tsinghua University's Yao Class and Peking University's Yuanpei Class to establish RoboParty Lab, dedicated to advancing full-stack software-hardware capabilities.

Next Step: Agentic Humanoid

RoboParty aims to lower the barriers to real-world robot development and deployment through fully open-source bipedal humanoid robots, uniting developers, robotics labs, and industry model clients within a unified ecosystem. The open-source hardware expands real-world touchpoints, the developer ecosystem amplifies task execution and data flywheels, and Party OS accumulates physical interaction experience that can be converted into transferable, generalizable intelligent capabilities.

Based on the latest milestone achievements from RoboParty Lab, RoboParty has established a four-tier technical architecture spanning from Loco-manipulation to Agentic Humanoid: At the Humanoid Locomotion layer, the team continues to refine data generation pipelines, motion retargeting algorithms, training frameworks, and Sim2Real / Real2Sim deployment workflows, while advancing BFM development. At the Humanoid Perceptive Interaction layer, focus is placed on environmental perception, human-scene interaction (HSI), and human-object interaction (HOI) to enhance robots' perception, decision-making, and continuous interaction capabilities in complex real-world environments. At the Humanoid Whole-Body Manipulation layer, BFM, VLA (Vision-Language-Action) models, and World Models are further integrated to enable coordinated full-body control and complex task execution, building a manipulation intelligence architecture with strong generalization capabilities. At the higher Agentic Humanoid level, the team explores task planning, autonomous execution, and skill composition based on the Agent + Skills framework, driving humanoid robots to evolve from isolated point capabilities toward fully autonomous generalizable intelligent agents.

Building on this technical system, RoboParty has recently open-sourced three core technical tools: MimicLite, UFO, and Human-to-Humanoid Tools. MimicLite delivers SONIC-level general motion tracking strategies using only 1/500 of SONIC's training compute resources, capable of training in just 2 hours on 8 H200 GPUs, while supporting rapid real-machine adaptation across different framework architectures. UFO provides an unsupervised reinforcement learning control framework compatible with diverse robot platforms, enabling low-cost training for general motion, highly dynamic maneuvers, and real-robot teleoperation. hhtools compresses complex Human-to-Humanoid motion retargeting workflows down to 30 seconds, supporting mainstream motion data formats, standard URDF robots, and Robot-to-Robot motion transfer.

Following this financing round, RoboParty will continue accelerating the mass production ramp-up of its open-source bipedal humanoid robots, enhancing full-system reliability, motion performance, mass production consistency, and cost competitiveness. The company will also refine the Party OS ecosystem and sustain heavy R&D investment in BFM and HFM technologies to maintain state-of-the-art performance.

Zheng Xuanle, Founder and CEO of Lightsource Capital, stated: "Embodied intelligence is a long-term industrial competition that tests not only technical judgment but also capital endurance and organizational evolution. Lightsource Capital is honored to have deeply participated in RoboParty's journey since its inception, leveraging the synergistic value of our three core roles: incubator, investor, and financial advisor. As an incubator, we support the company in defining its strategic direction and organizational structure; as an investor, we fund critical technical R&D during key growth stages; and as a financial advisor, we continuously connect RoboParty with top-tier capital and industrial resources. We are extremely excited to witness RoboParty, under the leadership of Huang Yi, rapidly build a platform where hardware, models, data, and developer ecosystems evolve synergistically. Moving forward, Lightsource Capital will continue to act as a long-term shareholder and entrepreneurial partner, accompanying this young team as it grows into a globally leading open-source full-stack platform for the embodied intelligence industry."

RoboParty's next-generation humanoid robot, the RP1, is scheduled for official launch in Q4 2026 — stay tuned.