Summary of Key Highlights
At this year’s World Artificial Intelligence Conference (WAIC), embodied intelligence (robots that can perceive and act like humans) became a major focus. The “work capability” of robots in industrial and commercial settings has significantly improved, but there are still technical and delivery challenges before they can be widely adopted on a large scale. Companies are intensively working on models (the “brains” of robots) and data (the “learning materials”) to address issues with generalization capabilities. New applications in commercial scenarios have emerged (for example, robots making ice cream), which has also driven growth in the entire supply chain, leading to a surge in orders for related components.
Detailed Analysis
1. Robots Are Becoming More Capable, but There Are Still Hurdles Before Widespread Adoption
The robots displayed at this year’s conference were much more capable than those from last year: some were stacking battery cells on production lines, scanning codes, and others were moving boxes in automobile parts factories; some could even perform complex wiring assembly tasks. Experts from 3C manufacturing plants noted that the robots’ stability and operational flexibility had improved, with some being able to handle soft kiwifruits without damage (thanks to tactile sensors that adjust their grip strength).
However, widespread adoption in factories faces several obstacles:
- Lack of perception and generalization capabilities: Robots often struggle to adapt to new scenarios or different types of parts.
- Industrial-grade requirements: Factories need robots to operate continuously for 24 hours, but many current models do not meet these standards.
- Delivery challenges: Companies must integrate robots with existing production lines, a process that is complex and time-consuming.
To overcome these issues, companies are deploying robots in real-world environments to collect data, which they use to train their models and improve their adaptability.
2. Models and Data Are Crucial for Robot Intelligence
Robots need intelligent models and high-quality data to function effectively. This year, companies showcased “world-class models” that can predict potential risks in physical scenarios, reducing the need for trial and error. There is no single dominant technology; approaches like VLA (Visual Localization and Association) and world-class models are being combined for better results.
Data collection remains a challenge: Only a small portion of the data collected (e.g., 9 hours’ worth) may be useful (due to repetitive or ineffective actions). Companies are investing in data systems, developing specialized equipment, or collaborating with companies that produce tactile data. Some are even advocating for industry-wide data standards to ensure more efficient efforts.
3. New Commercial Applications Are Emerging, Turning Robots from Decorative Items into Practical Tools
Commercial robots, once mostly used for ceremonial purposes, are now finding practical applications:
- Bionic robots as hosts: The fully bionic Moya robot served as a bilingual host and conducted evening interviews at the conference.
- Robotic restaurants: Sharpa and DQ opened a robotic restaurant in Shanghai in August, where robots use existing equipment to make ice cream without any need for facility modifications.
- Handling fragile items: Robots like DaiMeng can safely handle kiwifruits and glass drinks, adjusting their grip strength based on tactile feedback.
These seemingly simple tasks are actually quite challenging for robots, as they need to accurately perceive the texture of objects and adjust their grip accordingly.
4. The Supply Chain Is Booming, with a Surge in Component Orders
The rise in embodied intelligence has benefited entire supply chains:
- Increased production: The annual production of humanoid robots is expected to exceed 100,000 units this year (compared to just 20,000 in 2025); Zhuoyide has already signed agreements for mass production.
- Surging component orders: Companies producing six-axis force sensors (essential for robot mobility) have received more orders than in the entire previous year; these sensors are now standard equipment for industrial robots, prompting companies to expand their production capacity in Guangdong.
- New players entering the market: Automobile parts manufacturers are developing robotic reducers and electronic skins, while companies specializing in dexterous hands are also researching and developing new technologies (although issues like shorter lifespans and overheating remain).
The influx of new players is leading to improved component specifications and lower costs, which is beneficial for the entire industry.
Conclusion
Embodied intelligence is at a critical stage of transition from prototype development to widespread adoption. While robots are becoming more capable, technical and logistical challenges must still be addressed. Models and data play a pivotal role in advancing robot capabilities, and innovative applications in commercial scenarios are opening up new possibilities. The maturation of the supply chain is also accelerating this process. In the next one to two years, we can expect to see more robots integrated into factories and everyday life.