At the “Physical AI and Embodied Intelligent Robot Innovation Ecosystem Exchange,” a special event of the China International Fair for Trade in Services held in Beijing on September 11, 2026, Shi Hui, Vice President of Shanghai Fourier Intelligence Co., Ltd., delivered a keynote speech. He articulated a vision for service robots that extends beyond mere functionality.
Shi Hui began by sharing insights from Fourier’s decade-long journey in rehabilitation and elderly care, combined with recent advances in embodied intelligence. Established in 2015, the company launched China's first commercial lower-limb exoskeleton in 2017 and has since developed over 30 rehabilitation robots. In 2020, it introduced the Intelligent Rehabilitation Harbor, a systematic solution. Simultaneously, it has been developing humanoid robots since 2019, evolving from the GR-1 to the third-generation model, along with the open-source N1 robot, the ActionNet dataset, and the wheeled dual-arm GRW robot unveiled earlier this year in Shanghai.
While these appear as separate product lines, they share a unified goal: exploring how robots can serve humanity more naturally, safely, and efficiently. The company identified early on that robots must not only execute commands but also recognize human states, perceive environmental changes, and respond appropriately. This led to a focus on proactive interactive AI, which means sensing, understanding, and collaborating better within defined safety parameters, not acting autonomously. This vision supports an interaction-centric embodied intelligence platform built on core components like integrated joints and multi-degree-of-freedom dexterous hands, progressing through system design to motion and cognitive intelligence.
Reflecting on the rapid evolution of the industry, Shi Hui noted that while the initial focus was on whether robots could move stably, attention has shifted to their real-world utility. However, visiting elderly care facilities has reinforced a crucial point: being merely “useful” is insufficient; service robots must become “trustworthy.” This trust encompasses four dimensions. Professional trust requires robots to integrate into existing workflows of doctors and therapists. Interaction trust involves understanding human language, actions, and feedback. Safety trust is critical, especially in medical settings, requiring clear boundaries during physical contact. Operational trust depends on a robot's ability to sustain operations, optimize, and maintain itself over time.
In rehabilitation and elderly care, the goal is to address the gap between what individuals want to do and what they can physically do. The company redefines rehabilitation as rebuilding the ability to participate in life and elderly care as preserving autonomy and dignity amidst changing capabilities. This perspective frames the company's mission as managing a person's functional ability across their lifespan. Early experiences show that no single device can solve the complete rehabilitation puzzle. A systems engineering approach is necessary, starting with assessing individual states, designing intervention strategies, and seamlessly integrating into care workflows, considering that some abilities can be trained while others require assistive compensation.
Drawing on a 2017 experiment with brain-computer interfaces (BCI), Shi Hui explained that while early technology was immature, recent advancements have renewed interest. BCI is now being re-evaluated not for the novelty of “mind control,” but as an additional channel for robots to understand human intent, potentially aiding in rehabilitation training. The company's approach remains consistent: mature technologies enter real-world scenarios while frontier concepts are validated before integration. BCI can be broken down into signal acquisition, intent decoding, and robot control.
As the Intelligent Rehabilitation Harbor evolves, it is broadening its scope. While the 2020 version focused on multi-category devices with an operational framework, 2025 marked the integration of humanoid robots, and 2026 saw the inclusion of BCI and other novel interaction methods for testing. These capabilities, while not yet final solutions, are continually tested and validated in familiar environments.
To move from rehabilitation to broader community and home settings, several technical hurdles must be overcome. These include a safe and compliant physical system, a natural interaction system for seeing and understanding people, an execution system capable of real-world actions, a “brain” for task planning, and a system for continuous learning and adaptation. To address the open-ended nature of the real world, the company has incubated a model company focused on world models and in-context learning. The future measure of robot intelligence will be its ability to learn quickly in new environments, using demonstrations and corrections to adapt safely over time.
Interaction is another key focus, with investments in physical interaction (vision, touch, force feedback) and emotional interaction (emotion recognition). Future explorations include immersive interaction through VR/AR teleoperation and neural pathways such as EEG, all designed to increase the ways robots understand humans.
Commercialization follows a progressive path, starting with professional medical settings where safety and evaluation standards are clear. The next step is expansion into elderly care institutions, which offer longer-term human-robot interaction. This extends to community settings and eventually to homes, which will require higher technological maturity. The transition towards autonomy will be gradual, starting with teleoperation and human-in-the-loop systems before increasing autonomy as safety and capabilities allow. Ultimately, Shi Hui emphasized that the care sector is a complex system no single company can master. It requires collaboration among hospitals, care providers, technical partners, and operational experts. Fourier hopes to contribute through its robotic platforms, rehabilitation expertise, and system integration, building the next generation of care services. The future goal is not to make robots more human-like, but to make them more understanding, adaptable, and helpful, serving as long-term support for human capability.
This was the core message of Shi Hui’s presentation. Thank you.
Comments