This Action Plan has been formulated to fully leverage Beijing’s advantages in innovation resources within the field of artificial intelligence, actively seize opportunities for the development of embodied intelligence, accelerate the deep integration of technological and industrial innovation in embodied intelligence, and cultivate new avenues for the development of artificial intelligence.
I. Guiding Principles
Aiming at the forefront of AI technology and seizing the critical period for the development of embodied intelligence innovation and industry, this plan takes as its main objective the creation of a globally influential hub for embodied intelligence innovation and a growth pole for industrial development. Building on existing strengths and strengthening coordination,innovate mechanisms, and take innovation-driven development, platform support, scenario-driven applications, and ecosystem optimization as key breakthrough points. We will stimulate the vitality of diverse innovation entities, enhance the performance of foundational software and hardware, strengthen generic support capabilities, resolve challenges in scenario-based application implementation, accelerate the creation of a vibrant industrial ecosystem, and transform the potential of scientific and technological innovation into momentum for high-quality development.
II. Development Goals
By 2027, original innovation capabilities will be significantly enhanced. We aim to achieve breakthroughs in no fewer than 100 key technologies related to embodied intelligence "cerebrum" and "cerebellum" models, embodied intelligence chips, and whole-body motion control. We will produce no fewer than 10 internationally leading software and hardware products, and the upstream and downstream industrial chains of embodied intelligence will be largely localized.Infrastructure will be gradually improved, with the establishment of a series of new research and innovation platforms for world modeling and simulation, data collection, pilot testing and validation, and open scenario testing. Unified standards for embodied data collection management and testing/validation will be established to support at least 100 innovation entities in conducting technological innovation and accelerating product iteration.The industry scale will be further expanded, fostering no fewer than 50 core enterprises across the industrial chain, developing no fewer than 50 mass-produced products, and achieving no fewer than 100 large-scale applications across three major scenarios: scientific research and education, industrial and commercial sectors, and personalized services. The total mass production volume will be the first to exceed 10,000 units, fostering a 100-billion-yuan industrial cluster.The industrial ecosystem will be continuously optimized, with the establishment of no fewer than two specialized industrial clusters for embodied intelligence, the creation of industry-education integration bases in the field of embodied intelligence, and the cultivation of an industrial ecosystem with international influence.
III. Key Tasks
(1) Leading the Frontier of Embodied Intelligence Hardware and Software Technologies
1. Breakthroughs in Multimodal Fusion Perception Technologies
Support universities and research institutes in collaborating with leading enterprises to study spatio-temporal synchronization and calibration technologies for multisensor data, and efficiently integrate data from different perception sources;Research cross-modal learning algorithms to enhance the interaction and complementarity of data across different modalities; develop algorithms for interactive perception, active perception, and multimodal data completion to achieve efficient, automated alignment of large-scale multimodal data; and explore efficient, robust methods for unified multimodal representation and fusion of vision, language, and action in embodied environments to enhance robots’ perception and understanding capabilities.
2. Develop a Large-Scale Model for Embodied Intelligence “Brain”
Building on the city’s existing AI large-scale model industry foundation, support various innovation entities to construct a highly versatile multimodal foundational large-scale model capable of handling arbitrary modal inputs and outputs, as well as multimodal understanding and generation.Based on the multimodal foundational large model, develop an embodied intelligence “brain” large model equipped with capabilities such as spatial object perception, autonomous environmental understanding, and complex task planning, enabling embodied intelligent robots to handle complex tasks, adapt to dynamic environments, and predict future states.Develop an embodied intelligence large model that integrates perception, cognition, decision-making, and control, enhancing robots’ core capabilities in scenario understanding, logical reasoning, task planning, behavioral control, human-robot interaction, and autonomous learning, thereby improving their adaptability and generalization capabilities across different entities, scenarios, and tasks.
3. Enhancing the Capabilities of Embodied Intelligence “Cerebellum” Skill Models
Strengthen the role of enterprises as the primary drivers of innovation, encourage companies to establish technology innovation centers, and promote joint development between technology providers and application partners to create specialized and general-purpose skill models for embodied tasks. This will expand the skill library of embodied intelligent robots and improve their dexterity in handling complex tasks. Construct autonomous decision-making models to enable real-time perception, understanding, and rapid decision-making by robots.Research continuous learning methods for embodied intelligence systems and “human-in-the-loop” hybrid learning methods to achieve continuous improvement of robot skill models and autonomous adaptation to the environment.
4. Improve Robot Motion Control Performance
Research whole-body motion control strategies for high-dynamic-range robots, develop breakthrough motion control algorithms that integrate model predictive control with reinforcement learning, improve the control accuracy and response speed of embodied intelligence systems, and achieve dynamic balance and adaptive adjustment for robots.Research technologies such as dual-arm coordination, hand-eye coordination, and brain-body coordination; establish data feedback loops and online learning mechanisms; enhance the robot’s motion flexibility and execution efficiency; and achieve precise control and generalized operation of different components such as robotic arms and dexterous hands. Develop a universal robot motion control algorithm framework; implement modular and standardized design of hardware and software interfaces; and enable the integration of the embodied intelligence “brain” into heterogeneous robotic platforms.
5. Strengthen Technological Innovation and Supply Capabilities for Core Components
Optimize precision machining processes to enhance technological innovation and supply capabilities for core components such as sensors, reducers, integrated joints, and end-effectors. Develop key technologies including high-strength wear-resistant materials, precision machining and assembly, high-speed lubrication, and heat dissipation to improve motor performance and extend service life.Develop high-torque, high-precision, high-dynamic-response, and highly reliable servo drive systems and intelligent integrated joints; develop adaptive control algorithms to improve load-carrying capacity. Develop multi-sensor, highly integrated universal end-effectors, as well as high-precision robotic arms and dexterous hand systems to enhance fine and dexterous manipulation capabilities. Research robot lightweighting technologies; develop lightweight, high-strength materials, flexible materials, and high-performance batteries to improve operational endurance.
6. Develop domestically produced high-performance embodied intelligence chips
Develop general-purpose, high-computing-power, and high-bandwidth intelligent control chips for entire systems, providing critical support for the development and application of various embodied intelligence systems. Proactively plan for high-performance AI large-model cloud inference chips, ultra-low-power edge control and computing chips, and brain-inspired chips with autonomous learning and cognitive decision-making capabilities. Create modular, general-purpose intelligent modules for end devices to enhance the intelligent performance and deployment efficiency of terminal equipment.Conduct system integration of domestically produced embodied intelligence chips, communication modules, embodied intelligence "cerebrum" and "cerebellum" models, and world model simulation platforms to enable the efficient deployment of embodied intelligence operating systems and software algorithms on embodied intelligence robots, thereby building a fully domesticated end-to-end software and hardware ecosystem.
(2) Accelerate the Construction of New Research and Innovation Platforms
7. Build an Embodied Intelligence World Model Simulation Platform
Focusing on four key aspects—controllability, interactivity, 4D generation, and the embedding of physical laws—research will be conducted on efficient, scalable, controllable, and interactive next-generation video generation models. By integrating physical laws and common sense, an embodied intelligence world model simulation platform will be constructed.We will develop models of the macro-level operational laws of the world that can effectively simulate and predict future real-world states with limited input information, thereby helping embodied intelligent robots make optimal decisions and take actions in complex dynamic environments. We will generate diverse synthetic training data to reduce reliance on real-world data collection and enhance robots’ perception, understanding, reasoning, and generalization capabilities.
8. Jointly Build a High-Quality Multimodal General-Purpose Embodied Data Collection Platform
Construct a high-fidelity, multimodal, integrated perception-interaction data generation platform and establish a virtual-physical hybrid embodied intelligence data collection and training ground to support the collection of dynamic interaction data for robots across various real-world scenarios and complex tasks.Establish unified standards for embodied data collection and management; build an embodied data cloud platform covering the entire process from collection, cleaning, and annotation to management and sharing; establish a mechanism for the proactive discovery and utilization of data throughout the “training-tuning-correction” process; and accelerate the construction of high-quality, multimodal general-purpose embodied intelligence datasets. Research and establish mechanisms for the joint operation and open sharing of embodied intelligence data.
9. Build Pilot and Verification Platforms for Embodied Intelligent Robots
Encourage entities with experience in intelligent manufacturing to build a series of open and shared pilot and verification platforms for embodied intelligent robots. To address comprehensive pilot service needs—including the design of core components and robot prototypes, flexible manufacturing, process optimization, and small-batch production—deploy 3D printing, machining,PCB (printed circuit board) processing, and custom component manufacturing facilities. Design and construct customized production equipment and tools, explore market-oriented operational models, accelerate the industrialization of scientific and technological achievements, and increase product iteration speed.
10. Establish Open Testing Platforms in Real-World Scenarios
Develop a unified testing and verification system and standards, explore the establishment of joint verification mechanisms, and enhance the efficiency and credibility of testing and verification.Construct the Haidian Park Robot Open Training Ground, establish a multi-scenario, multi-task open physical testing environment, research cross-scenario adaptation technologies that integrate virtual and real worlds, and create an interactive testing and verification platform capable of intelligent adversarial testing. This will reduce the difficulty of transferring from simulation environments to real-world scenarios and improve the consistency of testing and verification for embodied intelligence across both simulation and real-world environments.
(3) Promote "Embodied Intelligence+" Multi-Scenario Demonstration Applications
11. Expand the Scale of Implementation in Research and Education
Explore new promotion models for embodied intelligence robots, such as open-source initiatives, financial leasing, shared trials, competitions and exhibitions, and educational training, and prioritize the promotion and implementation of embodied intelligence robots in the fields of research and education.Encourage innovative enterprises to establish a number of joint laboratories and technology innovation centers with universities, research institutes, and academic institutions to collaboratively develop new algorithms and applications, drive the technological upgrading and iteration of embodied intelligent robots, and accelerate the commercialization of research outcomes.
12. Accelerate the Large-Scale Deployment in Industrial and Commercial Scenarios
Promote central-local coordination in scientific and technological innovation, and encourage central and state-owned enterprises in automotive production, electronics manufacturing,industrial welding, coal mining, commercial retail, warehousing, and logistics to take the lead in opening up a batch of application scenarios. This will foster deep integration and joint R&D between scenario providers and technology providers, accelerate the accumulation of industry data, and further enhance the task comprehension and autonomous execution capabilities of embodied intelligent robots in complex production tasks and hazardous operations such as sorting and assembly, packaging and quality inspection, welding, and painting. It will also accelerate the replacement of workers in dangerous, repetitive, and physically demanding positions by embodied intelligent robots.
13. Proactive Exploration of Personalized Application Services
Proactively deploy solutions in human-machine symbiotic environments such as home services, elderly care, and healthcare; research human-machine safety and deep trust mechanisms; establish a theory of human-machine interaction value alignment; explore autonomous task discovery and planning mechanisms; and develop embodied intelligent robots based on mutual trust between humans and machines.Undertake research and development of embodied intelligent robot products to develop personalized service solutions for emotional companionship, health monitoring, anomaly handling, mobility assistance, and smart household tasks, and promote the demonstration application of embodied intelligent robots in elderly care facilities.
(IV) Optimizing the Embodied Intelligence Industry Ecosystem
14. Building a Full-Stack Talent Pipeline
Leveraging high-level universities and research institutions, identify cutting-edge technology researchers globally, and strengthen the recruitment and cultivation of strategic scientists, leading talents, and young researchers. Promote the establishment of general education courses on embodied intelligence in universities and research institutes, cultivate embodied intelligence talent through a tiered approach covering “original innovation—integrated innovation—open innovation,” and establish a mechanism for training interdisciplinary professionals.Establish industry-education integration bases in the field of embodied intelligence, encourage innovative enterprises to collaborate with universities and research institutes on joint talent training, and accelerate the cultivation of engineering and technical talent urgently needed by enterprises.
15. Pursue High-Level Open Cooperation
Leverage the strengths of flagship events such as the Zhongguancun Forum and the Beijing Zhiyuan Conference to successfully organize the World Humanoid Robot “Conference and Competition,” creating an internationally influential platform for cooperation and exchange in embodied intelligence and attracting internationally renowned institutions to establish a presence in China.Encourage leading technology enterprises to collaborate on projects with top international universities, establish overseas R&D centers, and support high-quality international expansion by innovation entities to tap into global markets. Promote technology-oriented social organizations in the embodied intelligence sector to enhance the quality and efficiency of their services in areas such as standard formulation and promotion, as well as international exchange and cooperation, thereby elevating the industry’s overall technological advancement and global influence.
16. Strengthen Tiered Enterprise Cultivation Services
Improve the tiered enterprise cultivation mechanism and service system to foster a cohort of unicorn enterprises, technology-leading enterprises, national high-tech enterprises, and specialized, refined, distinctive, and innovative “Little Giant” enterprises in the embodied intelligence sector. Enhance coordination between national funds and municipal and district-level funds, leverage the guiding role of the municipal artificial intelligence industry investment fund and the robotics industry development investment fund, and mobilize social capital to increase early-stage hard technology investments in the embodied intelligence sector.Strengthen the integration of grant and investment mechanisms, as well as equity and debt financing, to build a comprehensive, multi-tiered technology-finance service ecosystem and promote new models of government-bank cooperation in technology finance.
17. Create Embodied Intelligence Industrial Clusters
Actively promote coordination between the municipal and district levels, optimize the spatial layout of the embodied intelligence industry, establish specialized industrial clusters for embodied intelligence, and enhance the construction of public infrastructure for production, pilot testing, and assembly facilities to provide spatial guarantees and supporting facilities for embodied intelligence manufacturing.Strengthen scientific and technological service support for the embodied intelligence industrial cluster, establish a number of benchmark incubators, and integrate the service support chain for the transformation of cutting-edge technological achievements, entrepreneurial incubation, and industrialization. This will facilitate the transformation and implementation of a range of cutting-edge hard technology achievements within the cluster and accelerate innovation and entrepreneurship among young scientists.
Strengthen coordination among ministries, municipalities, and districts; leverage the role of national-level embodied intelligence platforms; enhance resource coordination; and systematically advance technological breakthroughs, industrial development, and scenario construction. Strengthen top-level planning in industrial layout, spatial cluster development, and the implementation of major projects; make effective use of the policy framework; ensure accountability; and actively seek to secure national-level projects.Through project organization models such as innovation consortia, open challenges, and competitive selection mechanisms, we will encourage diverse entities to jointly build new research and innovation platforms, promote resource sharing and complementary strengths, and carry out collaborative innovation around cutting-edge embodied intelligence technologies and industrial applications to enhance enterprises’ technological innovation capabilities. We will strengthen the alignment of technology supply and application demand, conduct dynamic tracking and evaluation of the industry, optimize the coordinated allocation of resources, increase interdepartmental funding coordination, enhance the level of science and technology ethics governance and risk prevention and control, and accelerate the cultivation of embodied intelligence industrial clusters.
This plan shall take effect upon issuance and remain in force until December 31, 2027. Should relevant national or municipal policies be adjusted during the implementation period, the plan shall be implemented in accordance with the latest national and municipal policy provisions.
