Firewheel Yintu completes 30 million yuan seed+ round of financing, focusing on the active interaction of embodied intelligence
As embodied intelligence accelerates out of the lab and into the real world, industry competition is shifting from "whether a robot can complete a task" to "whether a robot has the ability to actively understand and intervene at the right time".
When it rains, it will not take the initiative to hand over an umbrella; when an elderly person gets up staggering, it will not step forward to support them in advance; when a child is about to accidentally eat jelly, pills or foreign objects, it will not dissuade them in time; when the user is down and silent alone, it will not take the initiative to approach for company. Most of today's robots are already able to see, listen and execute instructions, but they still struggle to continuously understand changes in people and the environment, judge when to intervene actively, when to stay silent, and what actions to take.
"The real challenge for robots is not just answering questions, but judging when to act," said Kong Weigang, Founder and CEO of Huolun Yingtu. Focusing on this direction, the company continues to develop an active interaction brain for embodied intelligence, promoting robots to evolve from "instruction response" to "scene response", and from "passive execution" to "active service".
Recently, Huolun Yingtu announced the completion of a 30 million yuan Seed+ round of financing, led by Gaoxin Capital, with participation from Huayuan Capital and Angel Association, and additional investment from existing shareholders Sunwoda Group and DNA FUND. This round of financing will be mainly used for the iteration of the active interaction brain, the construction of the active interaction data system in real scenarios, and the large-scale mass production and delivery of the "Xiaojing" intelligent terminal.
The Dilemma: Why Robots Struggle with Active Interaction
"Robots can understand what users say, but they don't know when to respond actively," Kong Weigang, Founder and CEO of Huolun Yingtu, believes that this is the key challenge for embodied intelligence to move into the real world at present.
In the past few years, with the rapid development of large models, robots have made remarkable progress in visual perception, language understanding and task execution. However, after entering open environments such as households, education and commercial services, the challenges faced by robots are no longer just "whether they can complete tasks", but whether they can understand the changes of people and the environment in complex scenarios, and judge what to do next.
A large number of real-world demands are not clearly expressed. Users may not take the initiative to speak when they need help, and there are often no clear instructions before risks occur. Robots can answer questions, but it is difficult for them to identify potential demands; they can execute tasks, but it is difficult for them to judge when to remind, when to intervene, and when to stay silent.
This means that for robots to move from experimental environments to real life, what they need is not only stronger perception and execution capabilities, but also the capabilities of understanding scenarios, predicting demands and making autonomous decisions. The key to competition among robots in the future will further move from "perceptual intelligence" and "generative intelligence" to "active interaction intelligence".
Based on this judgment, Huolun Yingtu independently develops an active interaction brain, building a capability closed loop of "continuous perception - scene understanding - demand prediction - active decision-making - embodied execution - feedback optimization". Its core is not to make robots express themselves more frequently actively, but to enable robots to judge when to help, how to help, and when not to disturb, on the basis of understanding the relationship between people and the environment.
Luo Zhipeng, Co-founder and CTO of the company, said: "Real active interaction requires the integration of vision, voice, environmental status, user behavior and long-term memory, to understand the user's status in continuously changing real scenarios, and quickly judge the best intervention timing. Robots should not only know when to take the initiative, but also know how to take the initiative, and more importantly, know when to stay silent."
Breaking the Dilemma: How the Active Interaction Brain Enables Robots to "Read the Timing"
The core of active interaction is not to make robots speak more frequently actively, but to let them understand "when to intervene". Behind this, a large amount of real and continuous interaction data is required for support.
Different from internet texts, public videos or simulation data, Huolun Yingtu focuses on first-person real-scene interaction data, and continuously records changes in vision, voice, environmental status and user behavior through the "Xiaojing" terminal, to build a data system for active interaction. "What the active interaction model learns is not just 'what happened', but also to understand 'why intervention should be done at this moment, whether the user is willing to accept it, and whether the action really solves the problem'," said Kong Weigang.
At the technical level, the active interaction brain of Huolun Yingtu adopts the PAMI (Proactive Adaptive Multimodal Interaction) architecture, which integrates continuous perception, dynamic memory, demand prediction, timing decision-making and action feedback. Compared with general large models that focus on understanding and generation, its emphasis is on understanding the real-time relationship between people and scenarios: enabling robots to continuously perceive changes in the real world, and provide services in the right way at the right time.
The premise for the model to understand scenarios and people is to enter real scenarios. Huolun Yingtu promotes industrialization with the path of "terminal implementation - data backflow - model iteration": the "Xiaojing" terminal verifies the active interaction capability in real scenarios such as households, and continuously feeds back data and models, forming a closed loop from product use to capability evolution.
Backing Support: The "Dual Engine" of Technical Geeks and Industry Veterans
The evolution of active interaction from technical breakthrough to large-scale implementation requires both underlying algorithm capabilities and industrialization experience.
Luo Zhipeng, Co-founder and CTO of Huolun Yingtu, graduated with a master's degree in Software Engineering from Peking University. He has nearly 10 years of experience in R&D and management of artificial intelligence algorithms, and has long been deeply engaged in the R&D of AI algorithms and intelligent systems, with recent focus on large models, multimodal intelligence and embodied intelligence. He previously served as an algorithm engineer at Microsoft, and as a partner and Vice President of Technology at an AI unicorn enterprise, leading the end-to-end construction of AI platforms and intelligent application systems from 0 to 1. He led his team to win more than 50 championships in international AI competitions including ACL, EMNLP, KDD, CVPR and NeurIPS.
Kong Weigang, Founder and CEO of the company, has 15 years of experience in operation and management of listed companies and the AI industry, with long-term focus on platform operation, commercialization of technology products and industrial resource coordination. In his view, active interaction capabilities should not only be valid in the laboratory, but also be accepted by users in real scenarios, and form a replicable and sustainable business closed loop.
In addition, the core team of the company has both top university scientific research backgrounds and industry experience in leading technology enterprises. Members have educational backgrounds from Peking University, Tsinghua University, Fudan University, Shanghai Jiao Tong University, Imperial College London and other universities, and have R&D experience in enterprises such as Microsoft, Alibaba and Tencent, covering key directions including artificial intelligence algorithms, multimodal large models, robot intelligence and software and hardware systems, forming a complete capability chain from model R&D and technical engineering to product implementation.
Anchor Positioning: The "Active Interaction Capability Base" for Embodied Intelligence
"The robot industry will form a highly specialized division of labor system," said Kong Weigang. "Ontology enterprises define the form and motion capability of robots, general models provide perception, understanding and generation capabilities, and scenario enterprises are responsible for customer connection and service delivery. Active interaction capabilities connect people and robots, as well as model capabilities and real scenario value."
From the data platforms in the internet era, to the operating systems in the mobile internet era, to the rapidly developing physical AI today, every upgrade of the computing platform requires new underlying capability support. For embodied intelligence, the fact that robots can "see, understand and execute" is only the starting point; more importantly, whether it can continuously understand the changes of people and the environment, judge when to intervene and how to provide services, and continuously optimize based on real feedback.
This is also the positioning of Huolun Yingtu: taking the active interaction brain as the core, to build a capability base that connects the model, data, terminal and scenario ecology. The company does not attempt to replace robot ontology vendors, general model providers or scenario service providers, but through standardized and adaptable active interaction capabilities, it helps different terminals to enter real scenarios such as households, education and commercial services faster.
With the continuous accumulation of real scenario data, continuous iteration of model capabilities, and access of more terminals and partners, active interaction is expected to become an important infrastructure for the large-scale application of embodied intelligence. Huolun Yingtu hopes that in this process, robots will evolve from "waiting for instructions" to "understanding demands and providing active services", and truly become intelligent partners that can accompany and collaborate for a long time.
About Huolun Yingtu
As the first cutting-edge technology enterprise in China focusing on active interaction embodied brains, Huolun Yingtu is deeply anchored in the national artificial intelligence development strategy and the layout of six future industries in Shanghai, accurately solves the core pain points commonly existing in AI such as passive response, lack of autonomous action capability, and no collaborative interaction capability, redefines the interaction logic between intelligent agents, people and scenarios, and leads the technological innovation and industrial upgrading in the embodied intelligence field.
Relying on the large-scale implementation and wide application of the self-developed intelligent hardware "Xiaojing", we have accumulated globally scarce first-person face-to-face embodied data — massive human-computer interaction and full-scene execution data in real scenarios, which provides solid data support for the rapid iteration of the active interaction large model and the embodied brain. "Xiaojing" has completed joint debugging of software and hardware. Relying on upstream strategic cooperation resources such as Anova, Ruixing Microelectronics, TI and Malata, as well as downstream channels such as Sam's Club, Aigele and distributors across the country, it has received orders for more than 20,000 units, achieving dual breakthroughs in technical implementation and business closed loop.
The company currently has 32 core team members, including 4 doctors and 8 masters. The core team comes from top domestic universities such as Tsinghua, Peking University, Shanghai Jiao Tong University and Fudan University, and has won more than 100 awards in the world's top AI competitions, holding a number of core invention patents. Centered in Shanghai, the company has built an R&D system covering the Chinese mainland, Hong Kong, Macao and Taiwan regions. The company has completed two rounds of financing totaling 40 million yuan, and is supported by many top institutions including Shanghai Angel Association and Gaoxin Capital.
In the future, Huolun Yingtu will continue to open up core technologies, and work with industry partners to build an open embodied intelligence industry ecology. With the goal of becoming the world's leading provider of active interaction embodied brains, we are committed to making every intelligent agent a partner that understands users, takes the initiative and can collaborate, contributing to the development of China's AI industry.