Summary of Key Highlights
At this year's WAIC conference, embodied intelligence (AI with a physical form) received its own dedicated exhibition area for the first time, becoming a core focus alongside artificial intelligence computing. Compared to last year, the number of participating companies increased from over 80 to more than 200, with over 300 robots in action—all of them operating dynamically (last year, most were static displays). The most significant changes include:
1. Robotic Forms: Robots have shifted from a focus on showcase-worthy humanoid designs to forms that are better adapted to specific scenarios. This includes industrial wheeled models, consumer-friendly mini-robots, and manned exoskeletons.
2. Robot Intelligence: The “brains” of robots have evolved from VLA models that simply memorized fixed sequences of actions to world models that can make predictions based on context. However, these technologies are still in the early stages of development and are far from being practically useful.
1. Forms: From Humanoid Dominance to Contextual Adaptation
Last year, WAIC featured around 150 humanoid robots performing acrobatics or making coffee. This year, humanoid robots are no longer the sole highlight. Manufacturers are now designing them based on specific use cases:
- Industrial Focus: Companies like Zhiyuan have developed wheeled models (e.g., G2Max) that can transport goods 24/7. Shizhihang demonstrated a robotic assembly line for car wiring, using a combination of a wheeled base and dual arms for efficient, continuous operations.
- Consumer-N oriented Mini-robots: Smaller, more affordable humanoid robots (e.g., Xiaobumi from Songyan Power) are designed to attract the consumer market. They can dance and even update mobile phones. Some mini-robots are being used in scientific competitions.
- Experimental Forms: New models can change their appearance according to the environment. For example, Yushu’s manned exoskeleton (price: 3.9 million RMB) can be configured for indoor use with wheeled legs or for outdoor use with four legs (suitable for navigating grass and steps). Zhujidi Power’s modular robots can be assembled into a “centaur” configuration, capable of carrying 30 kg and pulling tires.
In short, while humanoid forms were once the standard, today’s trend is to design robots that suit different application scenarios, making them more practical.
2. Robot Intelligence: From Memorizing Actions to Predicting Outcomes
The intelligence of robots is a crucial aspect of their performance. Last year, VLA models (visual-language-action) relied on fixed sequences; this year, companies are experimenting with world models that allow robots to predict the outcomes of their actions before executing them.
- Long-Term Tasks: Robots like those from Galaxy Universal can perform tasks like making breakfast (baking bread, pouring drinks) for 3-5 minutes without interruption, even if the environment changes (e.g., a cup is moved).
- Complex Operations: Robots from Xingchen Intelligence can flip eggs or tie knots. They don’t just memorize movements but use cognitive processes to plan their actions (e.g., they slide the egg to the edge of the pan before lifting it).
3. Key Challenges: Data Collection and Dexterous Hands
The main barriers to the widespread adoption of robots are data collection and dexterous hands:
- Data Shortage: Although world models reduce the need for large amounts of data, high-quality, real-world data remains essential. This year, WAIC saw the emergence of specialized data collection companies that help manufacturers gather data more efficiently (e.g., Yu树 and Leju collect data in-house).
- Dexterous Hands: Precise control of robot hands is critical for practical applications. Companies like Daxiao Robot have developed devices that can be easily used by non-technical personnel to collect data.
- The industry is still working on standardizing data formats to facilitate collaboration.
4. Industry Status: Progress, but Still in the Trial and Error Phase
Despite significant improvements (e.g., increased task durations and more complex operations), robots are far from being widely used:
- Most demonstrations at the conference were conducted under controlled conditions; real-world applications face many challenges.
- Retail, logistics, and industrial sectors are still years away from seeing practical robot solutions.
- Key technologies (robot architecture, data standards, dexterous hands) are not yet standardized, and companies are still in the process of development.
As an entrepreneur at the conference said, “Simple tasks like dancing or greeting guests still require flawless execution. Complex applications will take many more years to mature.” The excitement around embodied intelligence indicates progress, but there’s a long way to go before these technologies become widely available.
In summary, while there has been significant progress in robot technology, practical applications are still in the early stages.