贾奎博士出席亚洲金融论坛：分享具身智能叩响商业化之门的核心洞察
Recently, the 19th Asian Financial Forum opened in grand style in Hong Kong. Held under the theme “Joining Hands amid Change, Sharing Wins in the New Landscape”, the forum brought together 3,600 political, business, and financial leaders from more than 60 countries and regions.
At this gathering focused on global economic transformation, Dr. Kui Jia, founder of DexForce Technology and tenured professor at CUHK-Shenzhen attended as a representative of the embodied AI industry, revealing, from a frontier-technology perspective, the collaborative potential and fresh win-win opportunities of embodied AI in times of change.
Dr. Jia pointed out that at the crossing point of current technological iteration and industrial restructuring, embodied AI is becoming a core breakthrough direction in the field of artificial intelligence. Unlike traditional large language models that depend on static data for training, embodied AI emphasizes real-time interaction between the agent and its environment through a physical body, achieving closed-loop learning from multimodal perception to adaptive decision-making. This paradigm is not only propelling revolutions in robotics, autonomous driving, and intelligent manufacturing, but also offers industries a key pathway for resilience upgrading amid global supply chain restructuring and the tide of digitalization.
As a pioneer of the embodied AI industry, DexForce Technology has, since its founding, been committed to converting cutting-edge academic research into deployable, hardcore technology solutions. The company has pursued comprehensive industry-academia-research collaboration with institutions and teams such as MIT, Tsinghua University, CUHK-Shenzhen, the Technical University of Munich, and South China University of Technology, securing a series of breakthroughs in core areas including perception fusion, reinforcement learning, motion control, and embedded systems.
In 2025, DexForce successively released a video-learning series, GAT-Grasp gesture perception, the GS-World world model, and the EmbodiChain open-source model training toolchain, among other frontier technologies. These not only cover the real-data domain, but also realize a fully automated task pipeline stretching from 3D generative simulation data through sub-millimeter environment modeling to fully automated model training and transfer to real robots, enabling robots to autonomously optimize action sequences in unstructured environments and markedly improving task success rates and generalization performance.
On the deployment front, the solutions of DexForce Technology have gone deep into multiple high-value scenarios such as scientific research and education, intelligent manufacturing, and supermarket retail. Through its self-developed integrated software-hardware platform, the company has helped customers intelligently restructure their production scenarios, empowering 50+ niche industries, serving 1,000+ projects, and achieving a complete closed loop from technology to product to business.
Take supermarket retail as an example: in 2025, DexForce introduced product forms such as robot baristas, robot blind-box sellers, and robot supermarket clerks. The robot barista has already been delivered in more than a dozen cities including Guangzhou, Zhuhai, Xi’an, Zhengzhou, Chengdu, Wuxi, and Xinjiang; the robot supermarket clerk has completed onboarding training at a leading light-meal chain in Hong Kong and is about to go officially live.
Dr. Jia stressed that the leading position of DexForce Technology grows out of sustained investment in fundamental research fused with engineering innovation. The company’s team has published nearly 100 papers at top conferences (such as NeurIPS and ICRA) and laid out dozens of core technology patents, forming a complete innovation chain from algorithm prototype to product iteration.
Looking ahead, DexForce Technology will keep concentrating on breakthroughs in the underlying technologies of embodied AI and exploring new models of human-robot coexistence and cross-domain collaboration. In a dialogue at the forum, Dr. Jia suggested that the next milestone for embodied AI will be the realization of general physical intelligence: enabling machines to learn and adapt autonomously in complex, ever-changing environments, much as humans do.
Most recently, DexForce officially open-sourced EmbodiChain, the first embodied AI toolchain built on a generative simulation world model that can automatically train VLA models and successfully deploy them on real robots. It requires no real-world data, trains VLA models with 100% synthetic data, and can be deployed directly on real robots, achieving zero-shot sim-to-real transfer. DexForce is now working with research institutions and industry partners around the world to jointly build open standards and testing platforms, pushing intelligent technology toward a safer, more reliable, and more inclusive future.
