Summary of Key Points
This news article focuses on the embodied intelligent robots presented at the World Artificial Intelligence Conference (WAIC). The main highlights are as follows: Embodied intelligent robots have made significant progress in their capabilities, but there is a high degree of homogenization and they have not yet been widely deployed; the elderly care scenario holds great potential, yet it presents significant challenges due to strict safety requirements; current robots suffer from an intellectual flaw akin to a “brain in a jar” (lack of interaction with the physical world); technological breakthroughs are needed to address issues related to data integration and the consolidation of various technical stacks; Tencent is actively investing in this field through platforms, models, and real-world scenarios.
Detailed Analysis
1. Embodied Intelligent Robots: Rapid Skill Improvement, but Prominent Issues with Homogenization
The robots displayed at WAIC can perform tasks such as sorting, pouring tea, and playing table tennis, showing a significant improvement from their predecessors two years ago, which were either stationary or suspended. However, Tencent scientist Zhang Zhengyou noted that these robots look similar and can perform similar tasks, whether in terms of appearance (e.g., humanoid forms or robotic arms) or the application scenarios (mostly targeting industrial settings). There is a general search for widely applicable use cases, but no clear breakthrough has been found yet, so the field is still in its exploratory phase.
2. Elderly Care Scenario: A Promising Direction with Major Hurdles
Zhang Zhengyou emphasized the importance of focusing on elderly care applications, highlighting the extreme difficulty of this area. Robots must be completely safe when in direct physical contact with humans; for example, applying too much force while assisting an elderly person could cause injury. To overcome this challenge, robots need to possess tactile, force-sensing, and visual capabilities, all of which must work seamlessly together—tactile feedback should be instantaneous (within 1 millisecond), while visual feedback can be delayed (up to 30 milliseconds). Tencent’s “Xiao Liu Plan” robots have been piloted in nursing homes to identify issues and make improvements through real-world use.
3. The Current Shortcomings of Robots: An Intellectual Limitation
Zhang Zhengyou described the current state of AI as resembling a “brain in a jar”—an extremely intelligent system that can reason, write articles, and answer questions, but lacks the ability to interact with the physical world. For instance, such a system may understand that a cup is used for holding water but not know how to hold it without dropping it, or it may not adjust its actions if the cup’s position changes. True intelligence requires integrating language understanding, visual observation, spatial cognition, body control, and environmental feedback to enable flexible adaptation in changing environments.
4. The Path to Smarter Robots: Integrating Four Types of Data
Training smarter robots requires four types of data:
1. Online videos showing how others perform tasks;
2. First-person operation videos (e.g., recording the process of making tea);
3. Touch data collected using sensor-equipped gloves;
4. Data from remote control of robots.
Integrating this information is essential for robots to learn effectively. The technical approach for embodied intelligence is still evolving; while large language models like ChatGPT have mature techniques, embodied intelligence is still in the exploratory stage. Recent trends include visual-language-action models (VLA) and world models, but there is no consensus on how to implement them effectively.
5. Tencent’s Approach: Building the “Brain” and “Body,” with a Focus on Real-World Applications
Tencent launched the Tairos Open Platform for embodied intelligence last year, providing a general framework for robot development. This year at WAIC, they introduced new models and agent frameworks such as Hy-Embodied-VLM-1.0 to enhance the connection between a robot’s sensory abilities (touch, vision) and its decision-making capabilities. Their “Xiao Liu Plan” robots in nursing homes serve as a platform for testing and refining technology in real-world scenarios—for example, how should robots respond if an elderly person suddenly falls? How can they provide safe assistance?
In summary, while embodied intelligent robots are still far from becoming common household items, the need in areas like elderly care is evident, and the direction for technological breakthroughs is clear: addressing safety issues, integrating data, and finding a unified technical approach will enable robots to become truly functional and useful.