Equipping the massage and physical therapy robot with a "super AI brain", "Embodied Storm" completes an angel round financing of over 10 million yuan | First released by 36Kr.
Written by | Wang Fangyu
Edited by | Su Jianxun
36Kr has learned that Guangdong Embodied Storm Robot Co., Ltd. (hereinafter referred to as "Embodied Storm") has recently completed an angel round of financing of over 10 million yuan. This round of investment is jointly invested by Xirong Shandao and Xinghe Capital, and Ren Chen Capital serves as the exclusive financial advisor. The raised funds will be mainly used for product research and development, mass production, and daily operations.
Embodied Storm was established in September 2023, focusing on the research and development and implementation of a new paradigm of embodied intelligent human-computer interaction, solving the problem of general human-computer interaction comfort through imitation learning based on massive real human feedback.
In March of this year, Embodied Storm received a seed round of financing from MiraclePlus.
The first product of Embodied Storm is an AI physiotherapy robot. According to the founder Li Xiaohui, the product has now achieved mass production and has been delivered in batches in the domestic and Japanese markets. Downstream customers include physiotherapy stores, beauty salons, and clinics, etc.
The AI physiotherapy robot of Embodied Storm is applied in a certain physiotherapy store in Japan Image source: Authorized by the enterprise
Physiotherapy robot is a medical device that combines robot technology with the principles of physical therapy. Currently, the demand for health management from consumers is growing rapidly, but the traditional physiotherapy industry is facing problems such as difficulty in recruitment, high labor costs, and high staff turnover. The physiotherapy robot can effectively solve these pain points.
Unlike traditional collaborative robots for physiotherapy, the AI physiotherapy robot of Embodied Storm equips the robot with an "AI brain", which can more accurately control the force and intensity of the operation.
For example, traditional physiotherapy robots generally only have several preset massage key gears such as mild, moderate, and severe, while the AI physiotherapy robot can adjust the massage force and direction in real-time according to the conversation interaction with the customer, providing a personalized physiotherapy experience.
The realization of the "AI brain" not only relies on the ability of the large model but also requires the vertical small model of the MOE (Mixture of Experts) architecture. The former can chat with customers, while the latter can accurately identify the specific instructions of customers for the massage service scene, jointly providing personalized and intelligent physiotherapy experiences for customers.
"Currently, it is very difficult for general robots to achieve universal, safe, and comfortable interaction with a wide range of people. The core reason is that individuals have different feelings about the comfort of physical touch. To solve this problem, Embodied Storm's strategy is to widely deploy robots in real scenarios to collect massive real human-computer interaction data." Li Xiaohui told 36Kr.
Li Xiaohui said that the AI physiotherapy robot of Embodied Storm has achieved a significantly lower cost than traditional competitors through multiple innovations. It can replace about 4/5 of the workload of manual physiotherapists and body care therapists with an average monthly salary of more than 10,000 yuan, and is welcomed by downstream customers.
"The six-dimensional force sensors of competitors are all purchased. Through self-research, Embodied Storm can achieve a cost less than 1/5 of that of competitors." Li Xiaohui said that Embodied Storm has achieved cost reduction for robot hardware through various methods such as hardware innovation and supply chain management.
The strategic goal of Embodied Storm is not only the physiotherapy robot but also a more general embodied intelligent robot.
The Embodied Storm team believes that a large-scale and low-cost real data source is the only solution for the breakthrough of embodied intelligent technology. It is collecting human-computer interaction data through physiotherapy robots in various places. The team will build an interaction database between multi-joint robots and a wide range of people (human bodies) by the end of 2025, which is expected to reach 5 million person-times, and mainly based on real human feedback data.
Different from the mainstream solution that uses imitation learning based on human teaching, which requires real humans to operate the robot for data collection, the imitation learning scheme based on human feedback adopted by Embodied Storm has achieved a significant cost reduction. Compared with traditional teaching data, human feedback can provide more accurate and targeted guidance in the learning process, helping the robot learn effective interaction strategies more quickly, and does not require humans to provide complete and accurate operation steps.
Embodied Storm is training a general human-computer safe and comfortable interaction model. The data it collects is centered on human feedback and intentionally decoupled from the robot form. The trained model is expected to empower various embodied intelligent robots in the industry.
In terms of the team, the "Embodied Storm" robot is jointly founded by several doctoral and professor teams from prestigious universities, including Li Xiaohui, a doctoral graduate in system control from the University of New South Wales, Mo Jixue, a doctoral graduate who served as an associate researcher at a well-known robot research institution and is now the CTO, and Guo Wenyi, a doctoral graduate in psychology who serves as the product director, and other technical experts. They have rich research experience in the fields of AI, robots, and human-computer interaction.