Summary of Key Points
At the World Artificial Intelligence Conference, a group of robotics startups that are not yet publicly traded but have a combined valuation of over 100 billion yuan made a significant appearance. Their robots have evolved from being mere showpieces to practical tools capable of performing real tasks in various scenarios such as factories, logistics, and pharmacies. These companies are experiencing robust fundraising (with single financings often starting at over 1 billion yuan), with state-owned capital playing an active role. However, they also face challenges such as a lack of authentic data and inadequate voice interaction capabilities. In the future, they will need to gradually break through towards commercialization by focusing on scenarios with high tolerance for errors.
1. Robots Have Truly Become Productive Tools
Robots have moved beyond being decorative exhibits in exhibition halls; they can now perform actual work. For example:
- Zhiyuan Robotics: Transferred the loading and unloading tasks from a flat-panel factory to the conference venue. The robots collaborating with JD Logistics can move boxes and stack them 24/7, eliminating the need for manual shifts and late nights.
- Ant Lingbo: Together with pharmacies, they have implemented a complete process for robots to pick up and deliver medications. Three robots share a central “brain” to handle orders, pick up medicines, and deliver them, and this system has already been put into use in real stores.
- Lingchu Intelligence: Specializes in the production of optical modules, where robots can grasp parts with precision down to the sub-millimeter level, more accurately and quickly than humans.
- Jizhijia: Set up a micro-production line, with humanoid and mobile robots working together to pick items and move boxes. They can even understand complex tasks like stacking and handling obstacles, almost without any pause.
These robots are no longer just toys; they represent a shift in the industry from technological exploration to creating real commercial value.
2. Behind the Billion-Dollar Valuations: Robotic Companies Are Seeing Explosive Financing
Most of the companies at the conference were unicorns (with valuations exceeding 1 billion US dollars), and their total valuation amounted to over 100 billion yuan. The financings were particularly substantial:
- Xingdong Jiye: Raised 2.5 billion yuan in just two months, with investors including state-owned funds such as Chengtong Fund and Jiangxi Guokong.
- Lingchu Intelligence: Has raised a total of 2 billion yuan, with additional support from “national team” capital.
- Zhi Pingfang: Received 5 billion yuan in its Series B financing.
In the first half of this year, these unicorns with valuations in the tens of billions of yuan raised over 40 billion yuan, accounting for 40% of the total financing in the field of embodied intelligence. The reasons for this surge include the empowerment of robot “brains” by large-scale modeling technologies, policy support, and the expectation of some companies going public (IPOs), which has attracted a large amount of investment.
3. Fast Progress, but There Are Still Challenges
Despite significant progress, robots still have several issues that need to be addressed:
- Advances: They can now understand data more deeply, and their dexterous hands can perform stable grasping tasks in industrial settings. Jizhijia’s models enable robots to “understand” physical changes in objects (such as how they stack or move), allowing them to know how to grasp them with just one glance.
- Challenges: Voice interaction is a major weakness. For instance, the robots at Beijing Quanjia Convenience Stores require microphones for communication, and their command understanding is slow and prone to errors. In noisy environments like conferences, they have difficulty hearing human voices. This is because most companies focus on vision and motion control, neglecting voice interaction, which traditional models struggle to handle in real-world scenarios with multiple people present.
4. The Biggest Bottleneck: A Lack of Real-World Data
For robots to become more intelligent, what’s most lacking is data from the real physical world:
- Data Mismatch: Internet videos show fantastical scenes, but robots need to understand real physical principles such as gravity, pressure, and temperature changes, which are not reflected in online data.
- Scarcity of Real Data: The amount of data in the physical world is much smaller than that on the internet, and simulations cannot accurately replicate factors like friction and deformation (e.g., how boxes bend under pressure).
As a result, companies are racing to establish “robot training centers” to collect real-world data. Some aim to collect millions of hours of useful data this year, while others use financing specifically for purchasing equipment to record the actual operations of their robots. The company that gathers the highest quality data will have smarter robot “brains.”
5. How to Move Faster in the Future?
The commercialization of embodied intelligence has not yet found a fixed path. Industry experts suggest:
- Don’t Set Directions Too Early: Just like OpenAI tried many products before creating ChatGPT, robotic companies need to explore various scenarios.
- Start with High-Tolerance for Errors: Focus on scenarios where robot failures have minimal impact, such as factories and warehouses. Once these areas are mastered, they can then expand into commercial (e.g., convenience stores) and everyday life applications.
In summary, the robotics industry has moved from being a concept to becoming a practical technology. However, to truly integrate into daily life, companies still need to solve issues related to data and interaction. This process requires a step-by-step approach.
(The entire text uses clear language that non-experts can understand, explaining the current state of the robotics industry, its financing situation, challenges, and future directions, with specific examples supporting each point.)