Embodied intelligence startup Spirit AI has raised nearly 200 million RMB in an angel round, with Oasis Capital making consecutive follow-on investments | Oasis Vitality

Spirit AI, the only embodied-intelligence company in China with full-stack AI-plus-robot productivity capabilities, recently closed nearly 200 million RMB in seed and angel funding. The angel round was led by HLC, with Fortune Venture Capital and Qianzheng Capital participating. Shunwei Capital and Oasis Capital, the company's seed-round investors, also continued to increase their stakes.

Founded in February 2024, Spirit AI secured heavy backing from multiple top-tier investors within just four months — a clear signal of the capital market's strong recognition of scarce talent in the embodied-intelligence track. Following this round, Spirit AI will continue to push forward on technology R&D and scenario deployment, building a general-purpose intelligent robot platform capable of handling diverse tasks and ushering in a new era for robotics.

Fengtao Han, Founder and CEO of Spirit AI, said: "Oasis has exceptionally deep thinking about entrepreneurship and life, and their professional investment team has been tremendously valuable to us. We look forward to building the new era of embodied intelligence together with Oasis."

An investment lead at Oasis Capital commented: "As a seed-round investor in Spirit AI, Oasis Capital has witnessed the company's rapid growth and iteration in a short period. Spirit AI has a team that combines both deep industrial understanding and cutting-edge technical expertise. The company's breakthroughs in pre-trained foundation models and embodied large models provide the foundation for building a truly general embodied-intelligence brain. Oasis Capital believes Spirit AI will continue to advance the development of embodied-intelligence technology. We will steadfastly accompany the company through continuous innovation and progress, creating the future of embodied intelligence together."

A World-Class Team, Full-Stack Capabilities in Place

Spirit AI has assembled a globally scarce elite team with leading capabilities across embodied large models, robotics, and scenario deployment. Team members hail from top institutions and companies including UC Berkeley, CMU, Tsinghua University, Peking University, Zhejiang University, Xiaomi, Huawei, Tencent, and DJI. The company possesses industry-leading embodied large-model technology and outstanding robotics R&D capabilities, with full-stack AI engineering expertise and core technical strengths in pre-trained models, imitation learning, and reinforcement learning. The team also holds leading positions in system-level robot design and optimization, functional safety technology, modular control software architecture, and high-performance robot control.

Fengtao Han, Founder and CEO of Spirit AI, has over a decade of experience in robotics. He previously served as Co-founder and CTO of珞石机器人 (Robostar), where he was a pioneer in China's high-performance lightweight industrial robots and the first to mass-produce force-controlled collaborative robots domestically. He led his team to successfully deliver more than 20,000 units across dozens of product models. Han also drove commercialization across 20-plus industries and 100-plus scenarios, achieving results for over 1,000 customers — demonstrating exceptional capabilities in productization and engineering.

In 2024, Han fully committed to embodied intelligence, bringing his world-class motion-control algorithms and productivity-grade robot design experience to launch his second entrepreneurial venture — choosing embodied intelligence, a new track where the world starts from the same line.

Yang Gao, Co-founder of Spirit AI, is an Assistant Professor at Tsinghua University's Institute for Interdisciplinary Information Sciences, Director of the Tsinghua Vision and Embodied Intelligence Lab, and a doctoral advisor. He has ten years of research experience in embodied intelligence, computer vision, and machine learning. Professor Gao studied under international computer vision master Trevor Darrell and has collaborated extensively with leading reinforcement learning scholars Pieter Abbeel and Sergey Levine.

Professor Gao has produced fruitful research across all three layers of the embodied-intelligence model. In reinforcement learning, for instance, he proposed EfficientZero and EfficientZero v2 — the most sample-efficient reinforcement learning algorithms in the world to date, highly praised by OpenAI Co-founder and reinforcement learning leader John Schulman. His years of research have focused on building a general embodied-intelligence brain capable of executing arbitrary tasks in any scenario, with frequent breakthroughs across internet video pre-training, imitation learning, physical-world reinforcement learning, and other embodied-intelligence domains.

With the vision of "building a world-class embodied-intelligence giant in China," Professor Gao decided to join Spirit AI as Co-founder.

Solid Technical Foundation, Crossing Intelligence Boundaries

Spirit AI has built distinctive technical advantages in both AI systems and robot hardware platforms.

In AI systems, Gao brings deep experience across multiple embodied-intelligence data sources including internet video pre-training, Sim2Real, imitation learning, and physical-world reinforcement learning — possessing "full-stack" embodied-intelligence data utilization capabilities. This full-stack data capability is essential for building a truly general embodied-intelligence brain.

Effective data utilization is the most critical factor in constructing embodied intelligence, and Gao has made important breakthroughs in this area. In key technologies including sample-efficient reinforcement learning and high-efficiency imitation learning, Gao is the designer of the world's most efficient algorithms. His EfficientImitate high-performance imitation learning algorithm achieves a 600% performance improvement over Stanford's VMAIL algorithm, enabling the highest generalization for imitation learning given the same data volume.

His high sample-efficiency reinforcement learning algorithm EfficientZero received high praise from OpenAI Chief Scientist for Reinforcement Learning John Schulman.

In efficiently leveraging VLMs for robot task understanding and planning, Professor Gao proposed the ViLa and CoPa models. CoPa was the first to demonstrate task-level generalization in VLM models, substantially outperforming VoxPoser proposed by Stanford's Fei-Fei Li. Notably, the world-leading Figure01 robot employs Professor Gao's published ViLa model in its top-level planning component.

Additionally, Professor Gao has explored new methods for training robot models using internet video. His team's recently proposed Any-point Trajectory Model (ATM) framework requires only minimal annotated data for training and demonstrates strong robustness, achieving up to 500% performance improvement over UC Berkeley's Pieter Abbeel's UniPi algorithm in long-horizon tasks. The ATM framework provides new theoretical support for few-shot and cross-embodied robot learning, substantially broadening the scope of usable data sources. This research was accepted by the top international robotics conference RSS 2024 and received perfect scores from all reviewers.

Robot hardware platforms determine the performance ceiling of embodied-intelligence systems, and hardware systems that balance high performance with low cost are one of the decisive factors for whether embodied-intelligence companies can achieve commercial viability. Spirit AI deeply understands the importance of co-optimizing hardware and AI system design. In robot hardware capabilities, Spirit AI possesses world-class robot motion control systems and hardware development capabilities, with deep accumulation in model predictive control (MPC), full-perception biomimetic compliance control, dynamic environment manipulation and interaction, high-precision collaborative control, real-time dynamic trajectory tracking, and flexible vibration suppression.

The team also has leading system-level electromechanical system design capabilities, encompassing high-precision load and vibration simulation, lightweight high-rigidity topology optimization, thermal-fluid analysis, and thermal limit cycle optimization.

Strong Deployment Capability, Promising Future Applications

With its unlimited scenario potential, embodied intelligence is poised to drive sweeping transformation from industry to services to home applications — heralding a trillion-level market waiting to be developed.

Upholding the ambitious vision of "creating the next generation of intelligent labor," Spirit AI will focus on commercial services and home applications. Leveraging its industry-leading embodied large-model technology and outstanding robotics R&D capabilities, the company will deeply understand customer needs, draw on extensive LTC and ITR process implementation experience, and complete an efficient commercial closed loop from technology R&D to product marketization — making general-purpose robots intimate partners for humanity and driving the world toward the intelligent robotics era. A new era where robots are as accessible as iPhones, available for anyone to purchase and use as they please, is approaching.

Appreciating Vitality

What do you think vitality is?

Vitality is courage, composure and decisiveness, constantly putting down roots and growing upward, tenacity and resilience.

— Fengtao Han, Founder & CEO, Spirit AI