Xspark AI Secures Nearly RMB 100 Million in Angel Funding to Accelerate the Commercialization of Physical AI

ME AI message, Xspark AI has completed its first round of nearly RMB 100 million in seed funding, led jointly by DHVC, Chuxin Capital, and SEE Fund Infinite Fund. Recently, trusted physical intelligence company Xspark AI (WuJie ZhiHang) closed its first seed round of nearly RMB 100 million. The round was co-led by DHVC, Chuxin Capital, and SEE Fund Infinite Fund, with participation from multiple financial investors including Shenzhen Angel Mother Fund and Shuimu Alumni Seed Fund, as well as industry players such as Beiyang Hai Tang Fund and Texgroup Family Office. The funds will primarily be used for core technology R&D, product iteration, and large-scale deployment of Physical AI. This is another landmark investment reflecting capital’s continued focus on Physical AI. Large models have granted robots increasingly strong understanding and planning capabilities—but as they enter factories, homes, and commercial environments, a more practical challenge emerges: robots must not only “know how to do” tasks, but also “do them stably” and “do them reliably.” Bridging the gap between model capabilities and the real physical world has become the key battleground for the next phase of Physical AI—and this is precisely why investors continue to back Xspark AI. Xspark AI was co-founded by former NIO autonomous driving core executive Qi Xiong, Tsinghua University tenured associate professor Wenbo Ding, and HKU MMLab post-2000 PhD candidate Tianxing Chen. Professor Yao Mu from Shanghai Jiao Tong University and postdoctoral researcher Shoujie Li from Nanyang Technological University serve as long-term advisors. The team spans domains including autonomous driving, computer vision, world models, and robotics. Since its founding, the company has consistently focused on developing key technologies for trusted Physical AI. Its self-developed world model achieved top rankings on the WorldArena leaderboard. Team members have deeply contributed to building global mainstream embodied intelligence evaluation systems. Their research has garnered over 600 citations on Google Scholar, more than 3,000 stars on GitHub, over 800,000 model downloads, and partnerships with industry leaders including NVIDIA, AgiRobot, ByteDance, and Shengshu Technology. On the technical front, Xspark AI integrates multispectral tactile sensing, its proprietary AIGC data generation model, and an embodied intelligence evaluation system to build a collaborative architecture: “Slow Brain VLM — Fast Brain WAM — Spinal Cord VTA.” The goal is to close the full loop—from data generation and model training to real-world robot deployment—enabling Physical AI to transition from lab validation to large-scale real-world applications. Moving Beyond “Showy Competition”: Rebuilding Industry Logic Through Trusted Deployment Physical AI is rapidly gaining momentum. From world models and VLA to reinforcement learning, robots are continuously pushing the boundaries of capabilities demonstrated in videos. At the same time, the industry is entering another race: rather than constantly upgrading model parameters or demo effects, the real challenge is enabling robots to enter real-world environments and reliably, safely, and scalably complete tasks. The reason is straightforward: the real physical world is full of dynamic changes and countless long-tail scenarios; black-box decisions from large models lack safety guarantees and struggle to handle real-world uncertainties, easily leading to equipment failures or safety risks. For the industry, a more critical question is emerging: What core capabilities should companies prioritize to truly remove bottlenecks in scalable embodied intelligence deployment? This is precisely the direction Xspark AI’s team has been deeply exploring. This perspective stems from the team’s extensive experience across both industry and academia. Xspark AI’s founder and CEO, Qi Xiong, previously led NIO’s large model delivery team and oversaw the industry’s first million-scale deployment of world models—propelling world models from technical validation to large-scale commercialization. As one of China’s earliest automakers to propose and implement scalable world model technology, NIO’s experience shaped Xiong’s view on Physical AI’s evolution. In his view, following the trajectory of autonomous driving development, physical intelligent devices must continuously enter real-world environments and accumulate real operational data to create a data flywheel—and ultimately reach Physical AI’s “ChatGPT moment.” The primary prerequisite for scalable deployment is whether robots can safely and accurately complete human-robot collaborative tasks. If Xiong represents the industrial deployment perspective, co-founder and CTO Tianxing Chen has long been at the forefront of embodied intelligence model research and evaluation systems. As the lead author of the benchmark embodied intelligence evaluation systems RoboTwin and RoboDojo, and founder of the leading open-source embodied community Lumina, Chen’s research spans nearly all key phases of recent rapid advancements in embodied models. He has also contributed to model training and evaluation at teams like ByteDance Seed, continuously advancing embodied intelligence infrastructure and open-source ecosystems. “From RoboTwin 1.0 to RoboTwin 2.0, then to RoboDojo—I’ve witnessed embodied models evolve from early low-parameter imitation learning to pre-trained VLA models, followed by a surge of various WAMs. This journey has made me acutely aware that embodied intelligence is approaching real-world deployment far faster than expected.” Yet, through continuous involvement in model development and systematic evaluation, he has increasingly recognized that while model capabilities are rapidly advancing, robustness, safety, and controllability in real-world deployment remain critical challenges. “The closer we get to deployment, the less we can focus only on whether a model ‘looks like it can do it.’ I want to drive the construction of trustworthy, evaluable, and deployable embodied intelligence systems—so models can reliably and safely complete tasks in the real world.” A Five-Dimensional Trusted Architecture: Building the Intelligent Core for Robots Focusing on trusted Physical AI, Xspark AI has developed a trusted physical intelligence core system based on its proprietary multispectral tactile sensing technology and high-quality data foundation—structured as a collaborative “Slow Brain—Fast Brain—Spinal Cord” architecture. This system can be deployed independently or integrated modularly with existing robot models via plug-and-play interfaces to endow diverse robotic platforms with safe cognitive decision-making and reliable execution capabilities. In the team’s view, true trusted deployment does not rely solely on model capability—it requires a complete closed loop formed by model capability, safety constraints, operational safeguards, verifiable evidence, and continuous iteration. The X-Brain Thinking (Slow Brain, VLM) handles high-level cognition and task planning through hierarchical semantic understanding and global decision-making. The model is pre-trained on simulated data, AIGC-generated data, and real-world scenario data, augmented with multiple layers of safety alignment mechanisms that enable risk alerting, instruction auditing, and proactive intervention. It continuously improves its real-world understanding by feeding back corner case data. The X-Brain Manip (Fast Brain, WAM) focuses on physical execution. Leveraging proprietary high-quality data and tactile fusion capabilities combined with safety constraint logic, it enables robots to perform precise, dexterous, and controllable operations in complex environments while enhancing generalization across diverse tasks. As the underlying safety mechanism, X-Reflex (Spinal Cord, VTA) enables millisecond-level real-time responses. By fusing multispectral tactile data with multimodal inputs, it provides real-time fallback for upper-layer decisions—rapidly identifying risks, defining safety boundaries, and autonomously executing emergency stops, trajectory adjustments, or active avoidance maneuvers. All abnormal data is continuously collected and fed back to upper-layer models to form a self-improving data loop.Beyond the model, Xspark AI has also developed its proprietary X-Eval evaluation system, systematically assessing the model across multiple dimensions—including safety awareness, generalization capability, and operational reliability—to identify scenario requirements and mitigate deployment risks, providing quantifiable and verifiable capabilities to support the scalable deployment of robots. A rare dual-strength team in industry and research, building dual barriers in technology and engineering. Xspark AI boasts a founding team with rare industry-academia combined expertise. Core members come from top-tier institutions such as Tsinghua University, The University of Hong Kong, and Shanghai Jiao Tong University, combining cutting-edge scientific innovation with hands-on commercialization and mass-production experience. The team has established a comprehensive talent layout in areas including autonomous driving, world models, robotic touch, embodied intelligence algorithms, and infrastructure. Co-founder and CEO Xiong Qi has spent 15 years in AI algorithms and 8 years in the autonomous driving industry. He previously led NIO’s Perception Delivery Team, Large Model Delivery Team, and Functional Systems Team, fully participating in the R&D and commercial deployment of NIO’s intelligent driving large models. He possesses mature experience in mass-producing millions of intelligent driving products, excels in engineering the practical application of frontier AI technologies, and has proven capabilities in team management and commercial strategy. Co-founder and Chief Scientist Ding Wenbo is a Tenured Associate Professor and PhD supervisor at the Shenzhen International Graduate School of Tsinghua University, a National Young Talent, and the Principal Investigator of the National Key R&D Program (Youth). He has received honors including Tsinghua University’s Special Scholarship, Lin Feng Counselor Award, and Academic Newcomer Award. As a leading scholar in domestic machine touch research, he has published over 80 papers in top international journals and conferences such as Nature, Science, Cell sub-journals, TRO, TMech, CVPR, and ICLR, with over 10,000 Google Scholar citations. He is also the corresponding author of the world’s first Nature Sensors paper on tactile large models, providing core scientific support for the company’s foundational technologies in tactile perception and robotic interaction. Co-founder and CTO Chen Tianxing is a direct PhD candidate at HKU’s MMLab under Professor Luo Ping, a globally renowned computer vision scholar, with long-term focus on core algorithms and infrastructure for embodied intelligence. He has published over ten papers at top international conferences including ICML, CVPR, ICLR, and RSS, won multiple Best Paper Awards and top rankings in academic competitions, with over 1,000 paper citations. He was selected by Sequoia Capital China and MIT Technology Review China for “AI25 — 25 Under 25 AI Innovators,” and has received honors such as Shenzhen University’s Special Prize and CCF Outstanding Undergraduate Award. As the lead author of the global benchmark embodied intelligence evaluation systems RoboTwin and RoboDojo, and founder of the leading open-source embodied intelligence community Lumina, his open-source projects have accumulated nearly 20,000 GitHub Stars and were selected by Alibaba Cloud’s EAI-100 as one of the Top 10 Embodied Open-Source Projects of the Year, continuously advancing the open-source ecosystem for embodied intelligence. In addition, Xspark AI maintains long-term collaborations with multiple academic advisors to explore frontier directions such as robotic touch and multimodal perception. The advisory team includes Mu Yao, Assistant Professor (Tenure-track) at Shanghai Jiao Tong University, who has been selected for national young talent programs, Shanghai’s Overseas High-Level Youth Talent Program, Zhiyuan Young Scholar, and EAI-100 2025 Academic Rising Star; he has published over 50 papers in top international journals and conferences with over 4,000 Google Scholar citations. Also included is Li Shoujie, a postdoctoral researcher at Nanyang Technological University in Singapore and PhD graduate of Tsinghua University, who has published over 20 papers as first or corresponding author in authoritative international journals such as Nature Sensors, IEEE TRO, and Soft Robotics, providing continuous academic support for the company’s frontier R&D. Technological strength ranks among the global top tier; deepening ecosystem collaboration to accelerate industry empowerment. In terms of technological accumulation, Xspark AI has gradually established a comprehensive capability system covering world models, robotic touch, multimodal data, and industry evaluation frameworks. In the field of world models, the company’s proprietary world model has topped the WorldArena global authoritative ranking and achieved first place in visual quality and third overall globally in the latest evaluation in May 2026—placing its technical capabilities among the world’s top tier. In robotic touch technology, the team has also achieved breakthroughs. In January 2026, the team’s research on the multispectral tactile sensor SuperTac was published in Nature Sensors and selected as a cover feature; accompanied by its proprietary 850-million-parameter tactile large model DOVE, it further advances multimodal tactile perception capabilities. This marks China’s first research成果 published in Nature Sensors with the institution as the primary contributor. Regarding embodied intelligence data infrastructure, Xspark AI has built a high-quality multimodal Ego dataset encompassing multi-view observations and high-resolution tactile information, developed an industry-leading robotic simulation platform with tens of thousands of rigid-body resources and over a hundred robotic skills, and innovatively launched an AIGC data generation platform. It has accumulated over 100,000 hours of training data and more than 10,000 hours of commercial delivery data, continuously building a high-quality data asset barrier. Beyond models and data, Xspark AI continues to drive industry infrastructure development. Core team members have deeply participated in building mainstream global embodied intelligence evaluation systems; their work has accumulated over 600 Google Scholar citations, over 3,000 GitHub stars, and more than 800,000 downloads. The team also leads the establishment of unified industry model interface standards and opens a complete open-source toolchain to promote industry standardization and further lower the barriers to embodied intelligence R&D. Regarding this funding round, Xspark AI Co-founder and CEO Xiong Qi said: “This nearly RMB 100 million seed round reflects capital market recognition of our technological capabilities. We firmly believe that AI’s mission is to empower the physical world and enable human-machine coexistence. Trustworthiness and practical applicability remain the ultimate goals for embodied intelligence development. In the future, we will continue to deepen our full-stack trustworthy embodied intelligence technologies, refine foundational industry infrastructure, and drive embodied intelligence toward reliable and practical adoption across industries.” From investors’ perspective, as embodied intelligence transitions from technical validation to industrial deployment, “trustworthiness” has become a new competitive focal point—and a key investment rationale for this round. Wang Mingyu, Senior Partner at鼎晖VGC (DHVC), said: “Trustworthy security is a critical prerequisite for embodied intelligence to enter real-world scenarios at scale—and an unavoidable challenge for the industry today. We have consistently focused on teams willing to patiently refine this foundational system. We recognize Wujie Zhihang’s precise understanding of industry pain points and their deep commitment to building trustworthy precision as their core capability; this long-term vision resonates strongly with us. We hope the company continues to strengthen its technological moat, lead the industry in establishing standardized trustworthy systems, and steadily carve out its own growth trajectory.”Tian Jiangchuan, Partner at Chuxin Capital, believes that embodied intelligence is at a critical inflection point, transitioning from technical demonstrations to large-scale real-world applications. He stated: “Over the past few years, large models have driven rapid advancements in AI within the digital world. However, when AI enters the physical world, challenges become significantly more complex: real-world scenarios are highly dynamic, with frequent long-tail cases. Robots must not only ‘understand’ and ‘reason’ but also perform actions stably, safely, and precisely. One of the key bottlenecks in current industry adoption lies in the scarcity of high-quality real-world data—laboratory, simulation, and publicly available datasets struggle to capture long-term variations and extreme conditions in complex environments such as homes, industries, and services. Model capabilities cannot be fully achieved through one-time training alone; continuous evolution is essential. WuJie AI Aviation has identified this critical pain point: on one hand, it continuously accumulates multimodal data from real-world scenarios; on the other, it integrates simulation and generative data to help robots continually expand training samples and improve generalization. Meanwhile, the company has developed a comprehensive system spanning task understanding, action execution, and safety safeguards, enabling robots to not only comprehend and assess tasks but also execute precise operations reliably and respond promptly to unexpected risks. We have strong confidence in the team’s composite advantages in technological frontier, industry experience, and productization, and look forward to accompanying WuJie AI Aviation as a key driver in bringing AI into the physical world.” Ma Lin, Partner at SEE Fund, said: “WuJie AI Aviation is a rare, hybrid team driven equally by industry and academia, characterized by pragmatism and focus—an extremely valuable trait today. In the field of embodied intelligence and physical-world data, they demonstrate exceptional foresight and have positioned themselves strategically at critical junctures. SEE Fund specializes in early-stage investments in key sectors where cutting-edge technology meets national needs; over the past three years, we have made significant investments in artificial intelligence and embodied intelligence. The founding team of WuJie AI Aviation has been known to us for a long time—we recognize them as another team combining strong business execution with high technological vision. Embodied intelligence will be a long-term industrial trend still in its early stages. We look forward to accompanying WuJie AI Aviation as it roots itself in industry, grows steadily, and jointly drives trustworthy embodied intelligence from the lab into real-world applications, accelerating technology commercialization and enabling Physical AI to generate tangible industrial and societal value.” (Source: Ifnar)