To develop highly generalized intelligent robots, an AI company in China completed nearly 200 million yuan of angel round financing in four months.

Spirit AI, the only intelligent company with AI+ robot productivity full-stack technology capability in China, announced that it has recently completed nearly 200 million yuan of seed round+angel round financing.

Founded in February, 2024, Spirit AI Company has received a lot of money from several top capitals in just four months, which fully reflects the high recognition of the capital market for the scarce team of smart track. After this round of financing, Chihiro Intelligence will continue to make efforts in technology research and development and scene landing, and strive to build a universal intelligent robot platform capable of various tasks and create a new era of robotics.

Spirit AI builds a world-class universal humanoid robot and the next generation of large-scale models and learning algorithms through forward-looking product thinking, leading large-scale model technology and excellent robot research and development capabilities, so that universal robot partners can enter thousands of households and drive the world into the era of intelligent robots.

Luxury team entrepreneurship, full stack capacity

Spirit AI's scarce elite team around the world is leading in terms of large-scale models, robots and scene landing. The team members of the company come from UC Berkeley, CMU, Tsinghua University, Peking University, Zhejiang University, Xiaomi, Huawei, Tencent, DJI and other top universities and enterprises at home and abroad. Chihiro Intelligence has industry-leading large-scale model technology and excellent robot research and development capabilities, and has full-stack AI engineering capabilities. It has outstanding core technology advantages in pre-training models, imitation learning and reinforcement learning. At the same time, the team is also in the leading position in robot system-level design and optimization technology, robot functional safety technology, modular control software architecture technology and high-performance robot control technology.

Han Fengtao, founder and CEO of Chihiro Intelligence, has more than ten years of experience in the robot industry. He was the co-founder &CTO of Luoshi Robot, and was the leader of domestic high-performance light industrial robots and the first person in domestic force control collaborative mass production delivery. His team successfully delivered more than 20,000 dozens of products. At the same time, the team led by Han Fengtao has also landed in 20+ industries and 100+ scenes, realizing the commercialization of 1000+ customers, and has strong robot productization and engineering capabilities.

In 2024, Han Fengtao devoted himself to body intelligence, and with his world-class operation and control algorithm ability and productivity-level robot design experience, he started his passionate second venture in his life, and chose the new track with body intelligence at the same starting line.

Levin, co-founder of Chihiro Intelligence, assistant professor of Tsinghua University Interdisciplinary Information College, director of Tsinghua University Visual and Embodied Intelligence Laboratory and doctoral supervisor, has ten years of research experience in embodied intelligence, machine vision and machine learning. Professor Levin studied under Professor Trevor Darrell, an international computer vision master, and had in-depth cooperation with Professor Pieter Abbeel and Professor Sergey Levine, the most famous scholars in the field of reinforcement learning.

Professor Levin has made great achievements in the three-layer model of embodied intelligence. For example, in reinforcement learning, Levin proposed the reinforcement learning algorithms (Efficiency Zero and EfficientZero v2) with the highest sample efficiency so far in the world. The algorithm was highly praised by OpenAI CoFounder and John Schulman, the leader of reinforcement learning. For many years, its research focus has been on building a universal embodied intelligent brain that can perform any task in any scene, and it has made frequent achievements in many fields of embodied intelligence, such as Internet video pre-training, imitation learning, and physical world reinforcement learning.

Under the concept and vision of building China into a world-class intelligent giant, Professor Levin decided to join Spirit AI as co-founder.

Solid technical base, crossing the intelligent boundary

At present, humanoid robots generally face the common problems of weak generalization and limited interaction-reflected in their work, which means that humanoid robots can only work in relatively fixed environments such as workshops and factories, and their interaction with the surrounding environment is limited, and their responses are not accurate and sensitive enough.

If humanoid robots want to be as smart as people, they need a big model as a smart brain.

In order to realize the interactivity and generalization of robots, one of the core barriers of Chihiro intelligence is to build a highly generalized and universal robot brain. ViLa(Vision-and-Language Models) and CoPa(Constraints of Parts) models are used in the brain as the multi-modal models of the intelligent robot with body. CoPa shows the task-level generalization in VLM model for the first time, and its performance greatly surpasses the VoxPoser proposed by Professor Li Feifei of Stanford. In addition, it is worth noting that the world-leading Figure01 robot also uses the ViLa model to make the robot have the ability to understand daily scenes and have common sense of life-this model framework was proposed by Levin, the co-founder of Chihiro Intelligence.

In addition, in the process of building embodied intelligence, the most important factor is the effective use of data. Levin has made an important breakthrough in this field. Among the key technologies including high sample efficiency reinforcement learning and high efficiency imitation learning, Levin is the designer of the most efficient algorithm in the world. Compared with the VMAIL algorithm of Stanford, the EfficientImitate learning algorithm proposed by him can improve the performance by 600%. The algorithm can maximize the generalization of imitation learning with the same amount of data.

EfficientZero, a high sample efficiency reinforcement learning algorithm developed by him, was highly praised by John Schuman, chief scientist of OpenAI reinforcement learning.

As a result, Spirit AI has built a unique technical advantage in AI system and robot hardware platform. On the one hand, in the field of AI system, Levin has rich experience in various embodied intelligent data sources such as Internet video pre-training, sim2real, imitation learning, and physical world reinforcement learning, and has the "full stack" of embodied intelligent data utilization capabilities. This full-stack data capability is very important for building a truly universal embodied intelligent brain.

At the same time, the robot hardware platform determines the upper limit of the performance of the intelligent system with body, and the hardware system with high performance and low cost is one of the decisive factors for the intelligent enterprise with body to realize the closed-loop business. Spirit AI deeply understands the importance of joint optimization design of hardware system and AI system. In terms of robot hardware capability, Spirit AI has the world-class robot motion control system and hardware development capability, and has accumulated profound knowledge in model predictive control MPC, full-sensing bionic compliance control, dynamic environment operation and interaction, high-precision collaborative control, real-time dynamic trajectory tracking and flexible vibration suppression.

Super landing ability, future application can be expected

With its infinite scene imagination, embodied intelligence is about to lead a wide range of changes from industry to service to home application, indicating a trillion-dollar market that needs to be developed urgently.

Adhering to the grand vision of "creating the next generation of intelligent labor force", Spirit AI will focus on commercial services and home use. With the industry-leading large-scale model technology and excellent robot research and development capabilities, it will gain a deep insight into customer needs, and rely on profound experience in implementing LTC and ITR processes to complete an efficient business closed loop from technology research and development to product marketization, so that general robots can become close partners of human beings and drive the world to the era of intelligent robots. A new era in which robots can be purchased and used at will like iPhone is coming to us.