At the World Artificial Intelligence Conference in Shanghai, key figures from China’s robotics sector highlighted the pressing need for more comprehensive data and enhanced AI frameworks to improve robots’ ability to interact with complex physical environments.
- Robotics faces a data shortage compared to large language models.
- Closed-loop systems linking hardware, data, and real scenarios are vital.
- Integration of cognitive science with AI needed for embodied intelligence.
What happened
During the World Artificial Intelligence Conference (WAIC) in Shanghai, industry insiders disclosed the current shortcomings hindering China's robotics companies from advancing embodied AI. They specifically noted the lack of sufficient multi-modal data and sophisticated AI 'brains' necessary to enhance robots' interaction with the physical world.
Leaders from companies such as SenseTime and Tencent described the challenges in building a closed-loop system that integrates hardware, data, and real-world applications. Despite collecting large amounts of human demonstration data, robots have not been widely deployed in actual environments, limiting the availability of effective training data and slowing progress.
Why it matters
The development of embodied AI is crucial for producing next-generation humanoid robots capable of navigating and understanding their surroundings autonomously. The current gap between video training content and relevant real-world data hampers the ability of robots to develop effective world models that can support complex reasoning and physical interaction.
Chinese robotics firms are at a pivotal point where integrating insights from cognitive science and neuroscience alongside technical advancements in AI and hardware could accelerate breakthroughs. Such integration is essential to move beyond AI systems that only process language to those that combine vision, spatial awareness, and physical feedback into a unified intelligence.
What to watch next
Chinese companies like Ace Robotics and Tencent are actively working on models and platforms that unify generative, cognitive, and physical intelligence. Ace Robotics' Kairos 3.1 model and Tencent’s suite of embodied AI tools signal a strategic push towards closed-loop systems that can learn from mistakes and self-correct in real time.
Observers should monitor how effectively these innovations translate into wider deployment of robots across industries in China. The ability to scale and replicate real-world scenarios for data collection, alongside tighter integration of AI brains with robotic bodies, will be key indicators of progress in overcoming the current limitations.