Summary of Key Points
The embodied intelligence industry (in simple terms, the field of intelligent robots that can perceive their environment and perform physical actions like humans) is experiencing explosive growth in 2026: In the first seven months alone, the industry has raised over 75 billion yuan in funding—four times the amount for the entire year of 2025 and 22 times that of 2024. There are numerous industry events, with the World Robot Conference featuring 300 companies and 2,000 exhibits. Even companies that provide data to the industry (i.e., “data vendors”) are raising funds faster than those directly developing robots (25 data companies raised 17 billion yuan in just six months). However, the most critical gap in the industry is not the quantity of data but the lack of “effective and meaningful data”—there is no unified standard to determine which data can truly enhance robot capabilities and enable cross-scenario reuse. To address this issue, the organizers invited four experts from various aspects of the data ecosystem to discuss the topic of “what constitutes high-value data for embodied intelligence” and hosted a closed-door live broadcast for relevant stakeholders to participate in.
Detailed Breakdown
1. **How Popular is Embodied Intelligence Right Now? Money and Interest are Surging**
- Funding Boom: In the first seven months of 2026, the embodied intelligence sector raised over 75 billion yuan, four times more than in 2025 and 22 times more than in 2024. Capital is flowing into the industry at an unprecedented pace.
- High Industry Momentum: The recently concluded World Artificial Intelligence Conference (WAIC) has yet to cool down, and the World Robot Conference is about to begin, with 300 companies showcasing 2,000 robots, creating a vibrant atmosphere reminiscent of a “robot festival.”
- Data Vendors Lead the Way in Revenue: Companies that provide data (for collection and processing) are generating more revenue than those that build robots. The fact that 25 data companies raised 17 billion yuan in half a year indicates that data has become an essential resource for the industry.
2. **Why is “Effective Data” Such a Critical Problem?**
There is a common misconception in the industry that “1 million hours of data collection” represents significant progress for embodied intelligence, but this is not always the case: **The duration of data collection does not equate to useful training data*. For example, 1 million hours of robot behavior videos may only contain 100,000 hours of actual usable data for training. What the industry lacks is a unified standard for evaluating data quality:
- Can this data improve robot performance? (e.g., making autonomous driving more stable?)
- Can it be reused across different scenarios or robots? (e.g., can obstacle avoidance data from a vacuum cleaner be applied to industrial robots?)
- Is it worth investing in for the long term, as it may become obsolete quickly?
Without such a standard, even large amounts of data are of little use. Therefore, “effective data” is the biggest challenge facing the industry.
3. **Four Experts Provide Insights on Data Value**
The four experts invited cover all aspects of embodied intelligence data and address different challenges:
- Peng Pai (Kuwa Technology): Failed data can be more valuable. As someone who works on autonomous robots, he emphasizes that trajectory data that clearly shows failure scenarios is more useful than data showing smooth operations, as it helps robots learn from mistakes.
- Guo Junliang (Qianxun Intelligence): The quality of the data affects the potential of the models. He studies the role of various data sources (human videos, real robot demonstrations, simulation-generated data, and feedback from failed experiments). He discusses whether to focus on quantity or quality when resources are limited.
- Lian He (Guanglu Intelligence): Data must be part of a closed-loop system to be effective. He explains how combining human data (e.g., cooking videos), simulation data, and real-world deployment data can improve model accuracy. He also discusses how to evaluate the effectiveness of data.
- Zheng Yujing (Kexing Shikong): Data must meet three criteria before it can be used: proper formatting, accurate annotation, and compliance with regulations. He explains how to manage data from different types of robots (e.g., vacuum cleaners and industrial robots) for consistent model development.
4. **Three Critical Questions the Industry Is Facing**
The discussion directly addresses three key issues:
- Where to Obtain Data and How to Allocate Budgets?: Different data sources (human videos, remote operations, simulation-generated data, and feedback from failed experiments) have different uses. With a limited budget, how should resources be allocated?
- How Much of the Collected Data is Actually Useful?: Only a small portion of the collected data (e.g., a few minutes out of 1 million hours) makes it into the training set. What factors determine which data to keep, and how should costs be calculated (based on duration or quality)?
- Is Data a Barrier or a Shared Resource?: Can data from one type of robot (e.g., a vacuum cleaner) be used for another (e.g., a delivery robot)? Who owns the data (the client, the robot company, the data provider, or a third party)? If data can be sold, can it become a competitive advantage for companies?
5. **Who Is This Live Broadcast Suitable For and What Benefits Will It Provide?**
- Target Audience: Entrepreneurs and technicians working in embodied intelligence models, robot development, data collection/simulation; professionals in the robotics, AI hardware, and smart manufacturing sectors; technology investors, researchers, and media.
- Benefits of the Broadcast:
- Insights from top experts on which data can enhance robot performance.
- A clear understanding of the costs involved in obtaining useful data.
- Clarification of data ownership and reuse/transaction possibilities.
- Opportunities to connect with model companies, robotics companies, and investors.
The live broadcast will take place on August 17th from 7:00 PM to 9:00 PM via Tencent Meeting. Interested parties can register by scanning the provided code.