Robot Industry "Moving Towards High Collaboration": Horizon Joins Forces with Ecological Partners to Welcome a New Chapter of Flourishing Industry

In 2025, embodied intelligence will reach a key turning point in transitioning from the laboratory to large-scale implementation. Horizon, as China's largest computing platform for consumer robots, has launched over 100 smart products in collaboration with its ecological partners through its Diguo robot, connecting more than 100 upstream and downstream partners and over 100,000 developers. It has also deeply focused on the full-stack key technologies of embodied intelligence, officially releasing two major open-source models for embodied intelligence on December 8: the HoloMotion model for the embodied intelligence core and the HoloBrain model for the brain core. Upholding the mission of "providing brains for robots that are everywhere in the future," Horizon is not only a witness to this wave of robotic intelligence evolution but also a deep participant and enabler.

In this context, Horizon organized the "Intelligent Leap from Smart Vehicles to Robots" special forum at the 2025 Horizon Technology Ecosystem Conference held on December 8-9. The forum deeply explored the technological development and commercial implementation of embodied intelligence, focusing on the ecological potential of cross-domain technology empowerment. By gathering consensus through the special forum, Horizon aims to work together with ecosystem partners to embark on a new stage of large-scale application of robot technology.

In his opening speech at the thematic forum, Horizon Founder and CEO Yu Kai stated: "From the very first day of its establishment, Horizon's original intention has been to create the 'brain' for robots. If autonomous driving is the 'exam' that is about to be 'completed,' then the 'major exam' for robots is just beginning, with ten thousand tracks where each has the opportunity to produce a champion. Horizon and its subsidiary Digua Robot have consistently positioned themselves as 'enablers'—only creating weapons, not engaging in battle. We will continue to provide the underlying chips, toolchains, and development platforms to empower innovators across various industries and move forward with the entire ecosystem."

Horizon Founder and CEO Yu Kai

From theory to product, from scenario to ecosystem

Decoding the Evolutionary Path of Embodied Intelligence

During the keynote session, several industry pioneers shared cutting-edge insights and practical achievements related to the foundational theories, underlying technologies, and implementation results of embodied intelligence.

Embodied intelligence's ability to decompose and generalize complex tasks has long been a bottleneck preventing the advancement of robotic intelligence. To address this challenge, Zhao Mingguo, a researcher at Tsinghua University and the Chief Scientist of Accelerated Evolution, took the stage to analyze how humanoid robots can develop reactive intelligence based on vision-driven inputs, using the learning of soccer skills as a case study. He outlined the evolutionary roadmap from individual intelligence to team intelligence. He believes that reactive intelligence needs to shift to a more fundamental approach, specifically by adhering to "human principles," and he innovatively proposed the "VISION-MAPPING-ACTION" strategic model to enable robots to react agilely and proactively with predictive capabilities. "We have a very ambitious goal: by 2050, we hope to establish a humanoid robot soccer team that can defeat the World Cup champions of that year," Zhao Mingguo stated excitedly. Currently, Digua Robotics provides technical support for some products under Accelerated Evolution, and in the future, both parties will work together to promote the widespread adoption and large-scale implementation of embodied intelligence in multiple scenarios.

Tsinghua University Researcher, Chief Scientist of Accelerated Evolution Zhao Mingguo

"Making human life better" has always been the original intention and goal of Horizon Robotics, which coincides with Fourier, the general robotics company. Fourier's founder and CEO, Gu Jie, delivered a keynote speech on "Creating 'Human-Centric' AI Robots," sharing technical insights on advancing robots towards emotional companionship.

"Fourier has been dedicated to one mission for ten years: creating AI robots centered around humans. Robots should empower and serve people, becoming excellent partners in the future," said Gu Jie. Driven by this vision, Fourier adheres to a technology strategy that centers on interaction, building a full-stack technology platform covering core components, body design, and interactive intelligence. This platform aims to enhance robots' ability to achieve more vivid, intelligent, and flexible human-machine interactions. It enables real-time physical feedback that is "light as a feather, stable as a rock" and allows robots to "understand" human emotions and intentions through eye contact and gestures, breaking boundaries to possess emotional computing and expression capabilities. The deep integration of robots from body to brain is supported by a powerful computing platform. Horizon's D-GU Robot S600, as the "most powerful brain" of Fourier's robot products, supports multi-modal model deployment with a computing power of up to 560 TOPS, upgrading semantic understanding, emotional perception, and end-to-end decision-making, thereby assisting Fourier in implementing its human-centered embodied intelligence strategy."

Founder and CEO of Fourier, Gu Jie

At the forum, Dr. Yu Yinan, founder and CEO of Vbot, presented the "Vbot Super Robot Dog (Big Head BoBo)." Big Head BoBo is the world's first intelligent robot dog that doesn't require remote control. Yu Yinan stated, "We believe that robots can make life better, but more importantly, they need to first enter our daily lives." To truly integrate robots into households and everyday spaces, Big Head BoBo possesses three core capabilities.

In terms of ontological intelligence, the robotic dog can adapt to various complex terrains in real life, maintaining stable walking in scenarios such as ramps and slippery surfaces, with a battery life exceeding five hours. In terms of spatial intelligence, the robotic dog's head has two degrees of freedom, both laterally and longitudinally, which better expands its field of vision. Regarding agent intelligence, by understanding semantics and comprehending the real environment, it can achieve companionship roles such as "being with you, for you, and helping you," capable of carrying loads, filming on the go, and taking care of family members and pets. In terms of computing platform, Big Head BoBo uses the Sweet Potato S100P chip, with 128 TOPS of edge AI computing power and high computing power utilization rate, providing sufficient performance support for the robotic dog's real-time decision-making and motion planning in complex environments.

Vbot VitaPower Founder and CEO Yu Yinan

Actually, how to enable robots to possess human-like abilities in perception, planning, decision-making, and behavior has always been the core direction for breakthroughs in embodied intelligence technology. Horizon Robotics Lab, as Horizon's forward-looking research institution for embodied intelligence, with the lab head Su Zhizhong taking the stage to explain in detail Horizon's research layout and innovation in the key areas of embodied intelligence, such as simulation data engines, motion control, visual language navigation, and general systems.

In order for a robot to ultimately complete its tasks, it first needs to undergo data training. The EmbodiedGen embodied intelligence simulation data engine developed by the laboratory constructs diverse and physically accurate digital worlds using generative AI technology, providing an ideal simulation training ground for the robot's "perception-decision-execution" closed loop. After training in the simulation environment, the robot needs to be able to control its body effectively in the real world to accomplish motor control functions. The HoloMotion humanoid robot full-body motion control foundational model plays the role of the "human cerebellum," possessing real-time, robust, and generalizable full-body motion control capabilities. It also offers a complete pipeline from data acquisition to reinforcement learning multi-machine parallel training to real machine deployment. Currently, HoloMotion can achieve tracking of any trajectory. For the robot to reach its destination, it must have visual language navigation capabilities. In this regard, the robot laboratory has completed the FSR-VLN visual language navigation solution, with a model that has automatic inference capabilities, can receive high-level concise instructions, and can accurately locate corresponding objects in the environment. Finally, to reach the destination, the robot relies on its "hands" for dexterous manipulation to complete specific tasks. The HoloBrain general operation foundational model achieves breakthroughs through enhanced spatial perception and unified action space technologies, integrating real and simulated data to give embodied intelligence human-like spatial awareness and delicate operation. Models and algorithms alone are not enough; Horizon Robotics Laboratory has also developed the embodied intelligence framework RoboOrchard, which balances data quality, flexibility, and adaptability to multiple entities.

Based on these key technological breakthroughs, Su Zhizhong stated that Horizon Robotics Lab aims to use its exploration results to support the entire embodied intelligence industry, including companies like Digua Robotics. He emphasized, "The embodied intelligence industry is still in its early stages, with numerous technical challenges awaiting solutions. Everyone needs to collaborate to push the boundaries of technology."

Horizon Robotics Lab Director Su Zhizhong

Under the guidance of full-stack key technologies in embodied intelligence, Horizon Robotics has continuously achieved results in the field of robotics. Based on the vision of "becoming the mother ecosystem of the robotics era," its subsidiary Diguo Robots focuses on three major directions: "robot products that have been mass-produced," "ubiquitous new applications for robots," and "general embodied intelligent robots leading to the future." It provides comprehensive support for robot partners in development and large-scale deployment. Popular products equipped with the Diguo Robot computing platform have been launched successively.

Hu Chunxu, Vice President of Developer Ecosystem at Digua Robotics, took the stage to share practices and prospects: "We want to expand the ecosystem of robots, support more developers and enterprises, and help them turn imagination into real, tangible products." To achieve this expectation, Digua Robotics has built a comprehensive development infrastructure for developers, including a smart computing base integrating both software and hardware, and an open development platform that combines both edge and cloud. In his speech, Hu Chunxu introduced that the Digua Robotics smart computing base is centered around the BPU architecture, integrating multiple computing units to launch the Xuri series chips and RDK developer kits, with computing power ranging from 5TOPS to 560TOPS. Coupled with an open-source algorithm library of over 300 pre-trained models, it provides high-efficiency computing support for various robots, integrating both software and hardware.

Moreover, not long ago, Digua Robot officially launched its one-stop development platform. This platform integrates three major modules: data closed loop, embodied intelligence training ground, and Agent services. Developers can complete algorithm development and validation without complex programming, reducing the product cycle from idea to mass production to as fast as one year. Based on simpler end-to-end development infrastructure, Digua Robot has achieved a 180% year-on-year increase in shipments, doubled its customer base, and attracted over 100,000 developers from more than 20 countries worldwide, along with more than 500 partner companies, incubating over 5,000 open-source projects. Furthermore, by launching the "Gravity Plan," it provides developers with resources such as capital connections, supply chain assurance, and overseas channels, comprehensively accelerating the innovation landing process.

Vice President of Developer Ecosystem at Digua Robotics, Hu Chunxu

Famous people don't speak in riddles.

Roundtable Discussion on Consensus and Divergence in Embodied Intelligence

The embodied intelligence industry is at a critical stage of transitioning from technological breakthroughs to commercial implementation, with core topics such as technical routes, scenario adaptation, and ecosystem construction receiving significant attention from the industry. During the roundtable discussion hosted by GeekPark founder Zhang Peng, several guests from academia, research, and industry gathered to engage in a lively discussion on topics such as robot forms, technical routes, implementation paths, and the co-construction of the industry ecosystem. They reached a consensus to explore a feasible path for the development of the embodied intelligence industry.

The design concept of a universal base + functional modules has become a consensus. Zhang Di, the founder and CEO of Benmo Technology, believes that "the core difference between humanoid and wheeled robots lies in the approach chosen to solve the major issues of mobility and operation." In this regard, Shi Hailin, head of the AI large model and interaction department at UBTECH Research Institute, stated that the choice of robot form should be viewed through the lens of industrial application scenarios, and agreed with Zhang Di's proposed design concept of a universal base + functional modules, considering it beneficial for accelerating the implementation of embodied intelligence. "Using a universal platform at the base and gradually adding various functional modules to achieve robot universality is a faster path to drive embodied intelligence towards commercialization and industrialization," Zhang Di stated.

World models are key to the further evolution of robot intelligence. Huang Guan, founder and CEO of Extreme Vision, stated, "And it is not something that will be conquered in 5 or 10 years. Thanks to the development of generative AI, we have actually seen the dawn of its resolution." Huang Guan emphasized that the value of world models for embodied intelligence is comprehensive, whether in terms of data sources, learning paradigms, or model architecture, they all possess immense value.

The gradual progression of robot functions towards generalization is the core path for C-end commercialization. Zhang Di believes that consumer-grade robot products are currently on the eve of the "iPhone moment" and need to start with a single function, continuously adding innovative capabilities to achieve a transition from quantitative change to qualitative change. Wang Cong, CEO of Digua Robotics, agrees and states that "consumer electronics must still return to the most fundamental product needs, adding functions little by little," while also "finding specific scenarios to solve problems."

The industry will eventually move towards diverse scenarios and varied ecosystems, which is destined not to be a monopolistic market. Wang Cong stated that the future of the robotics industry is bound to be a non-standardized revolution where a hundred flowers bloom; robots will no longer be the stereotypical images from sci-fi movies but will deeply integrate into thousands of industries, serving real-world scenarios as intelligent terminals. He also stated that Digua Robotics, as a platform company, provides developers with support in computing power, algorithms, and toolchains, aiming to perfect the underlying infrastructure to truly accelerate the arrival of the future. "What kind of scene would make us genuinely happy? It's not because I've created an incredibly intelligent robot, but because I see a variety of consumer electronics, smart hardware in the home, and robots across different industries, all empowered by us. This is the image I envision, which would make me genuinely happy; this is our emotional side," Wang Cong said.

Returning to the core discussion of the "From Intelligent Driving to Robotic Intelligence Leap" forum, a clear consensus is: the development of embodied intelligence has progressed from single technical exploration to a new level of multi-dimensional collaboration in scenarios, data, and ecosystems. In the future, Horizon will continue to uphold the ecological philosophy of "open and win-win, full-dimensional altruism," and, driven by both technology and ecosystems, will work with partners to jointly promote the intelligent upgrade of automobiles and robots, making tomorrow's human life safer and better.

As Yu Kai said in his closing speech at the conference, "Everything we do is to empower machines, but the goal is not to make machines great, but to make humanity greater."