Hello! I'm your financial news analysis assistant. This news article about the departure of key members from Xiaomi's intelligent driving team to start their own businesses contains a wealth of information. It not only highlights the internal disputes over technical approaches at Xiaomi but also reveals the significant shift in the AI industry from a focus on "pure software" to applications in the "physical world."
To help you understand this easily, I will first provide a one-sentence summary of the core content and then break it down in five aspects using simple language.
Core Content Summary
Two of Xiaomi's top technical experts, both born in the 1990s (Wang Naiyan and Chen Long), have left the company due to internal conflicts over technical strategies and the desire for greater career opportunities. They have joined the booming field of "embodied intelligence," which involves using AI to control robots for practical tasks. This move indicates that the main battleground of AI competition is shifting from chatbots on phone screens to real-world physical operations.
---
In-Depth Analysis: Five Key Aspects to Understand This Departure
1. **Why Did They Leave?**
It's Not About Being Laid Off, but About Losing the Competition and the Changing Trends**
When executives leave a company, many assume it's because they were replaced. However, the situation is more complex:
- The Fierce Internal Competition: Xiaomi adopted a "dual-track" approach to intelligent driving: one focused on L2 (assisted driving, still requiring human supervision) led by Chen Guang, and the other on L3 (more advanced autonomous driving) led by Wang Naiyan.
- Current Reality: L2 technology is ready for mass production and generates direct revenue, while L3, despite its advanced capabilities (such as setting a record on the Nürburgring), is still far from being commercially viable and poses significant risks for car manufacturers.
- Decision: Resources were allocated to L2 because it was more profitable. Wang Naiyan felt that Xiaomi wasn't providing enough support to develop L3, which he described as a lack of opportunities for meaningful innovation.
2. **What is “Physical AI,” and Why is It More Challenging and Valuable?**
The term "Physical AI" refers to AI systems that can control physical objects in the real world:
- Current AI (e.g., DouBao, Siri): These are digital assistants that process information from the internet.
- Physical AI: They can control robots to perform tasks like lifting objects or driving vehicles.
- Challenges: Physical AI requires extensive data collection, as traditional AI relies on internet data. For example, learning how to handle objects safely involves practical experiments that are costly and time-consuming.
- Technical Difference: Physical AI needs to understand physical laws, such as the force required to lift an object without breaking it.
3. **Wang Naiyan's Approach: Teaching AI to Drive Without Imitating Humans**
Wang Naiyan, former CTO of TuSimple, advocates a unique approach called "representation learning":
- Traditional Methods: AI is trained by observing human driving videos, similar to teaching a child.
- Limitations: This approach only teaches what humans can do well, and it requires massive amounts of data.
- Wang Naiyan's Method: Instead, he feeds AI maps, traffic rules, and vehicle parameters to let it derive its own driving strategies.
- Success: This approach helped him set a record on the Nürburgring and represents a breakthrough in AI control.
- Implications for Entrepreneurship: If AI can master these physical control strategies, it could improve safety in extreme conditions.
4. **Chen Long's Approach: Giving Robots “Perception” Abilities**
Chen Long, from Wayve (a UK autonomous driving startup), focuses on perception and multi-modal interactions:
- Current Issues: Robots often lack accurate spatial understanding.
- Chen Long's Solution: He combines driving data with robot data for better perception.
- Business Focus: His company develops software for robots to perform various tasks efficiently.
5. **The Shift in the AI Industry**
The departure of these experts reflects a major shift in the AI industry:
- First Half (2023-2024): The focus was on large models that processed large amounts of data and required strong computing power.
- Second Half (2025 and Beyond): The focus is on practical applications in manufacturing, logistics, and transportation, where AI can replace human labor.
- Reasons for the Shift: AI models have matured, and capital is flowing into embodied intelligence due to its potential for massive economic value.
- Talent Flow: Top AI talents are leaving car companies for startups that offer more opportunities to explore cutting-edge technologies.
Summary for Laypeople
1. The Future of AI Lies Beyond Chatbots: The competition will move to industries where AI can replace human labor.
2. Xiaomi’s Intelligent Driving Team is Still Strong, but It’s Adjusting: Xiaomi is focusing on L2 for immediate benefits, which may limit technological innovation for enthusiasts.
3. Entrepreneurship Comes with Risks: Starting a business in physical AI is risky, but it can lead to significant breakthroughs.
One-Sentence Comment: Xiaomi’s “talented individuals” have chosen to pursue their own dreams in the emerging field of physical AI, pushing the boundaries of technology.