The debate over the future of autonomous driving technology has been ongoing for years. At the center of it lies the choice of sensors: LiDAR-based or vision-based systems.
China’s leading EV startup Xpeng announced that it will transition its autonomous driving system from LiDAR-centric to a vision-based approach, similar to Tesla. This shift is more than a mere technical update—it symbolizes a major paradigm change in the autonomous driving industry.
1. LiDAR vs Vision: What’s the Difference?
LiDAR sensors emit high-powered laser beams to precisely measure the distance and shape of surrounding objects. It functions like a “mechanical eye,” offering high-precision data and being less affected by weather or lighting conditions. However, the equipment is bulky and expensive, making it a cost driver for vehicles.
In contrast, vision-based autonomous driving uses cameras to “see” the surroundings like a human does, with AI making decisions. The core lies in cameras, computing power, and trained AI. The hardware setup is relatively simple, and as the technology matures, costs decrease while accuracy improves.
In short, while LiDAR is accurate, static, and mechanical, vision-based systems are adaptive and akin to an “AI brain” capable of learning.
2. What Has Changed in Xpeng’s Strategy?
Previously, Xpeng adhered to a sensor-fusion autonomous system that combined both LiDAR and cameras. This worked well for precise urban navigation but faced limitations in cost and scalability.
Recently, however, Xpeng adopted a pure vision-based system (XNet), aligning more closely with Tesla’s philosophy. This change includes:
- Removal of LiDAR: Reduces cost and simplifies hardware
- AI-centric architecture: Recognizes and interprets surroundings using only cameras
- Enhanced big data learning architecture
- Long-term goal of achieving “Pure Vision” autonomy
This marks a shift in Xpeng’s autonomous driving technology from being hardware-driven to software-driven.
3. Why Is Vision-Based Better?
Vision-based systems offer a self-improving structure over time:
- The more users, the more data is collected,
- AI learns from this data, enhancing decision-making capabilities,
- Software updates (OTA) apply improvements automatically to vehicles.
In other words, “the more it’s used, the smarter it gets.” Tesla has already leveraged this model, training on billions of kilometers of driving data and establishing a dominant position. Xpeng is signaling its intent to follow suit.
While LiDAR can also collect data,
→ it generates large volumes and is less efficient for learning,
→ and has limitations in enabling AI to perceive and judge like a human.
Thus, the true competitive edge in autonomous driving is shifting from sensor specs to AI learning capacity and data processing power.
4. Are Xpeng and Tesla Doing the Same Thing?
At first glance, both companies seem similar in adopting a vision-based autonomous driving approach. However, there are notable differences in their strategies:
| Category | Xpeng | Tesla |
|---|---|---|
| Sensors | Cameras + Ultrasonic + Radar | Camera-only system (no LiDAR or Radar) |
| Map Usage | Uses HD maps + Mapless driving capability | Fully vision-based, no maps used |
| AI Architecture | BEV (Bird's Eye View)-based XNet | Occupancy Network approach |
| Data Learning | Server-based learning + OTA updates | Real-time improvement via Shadow Mode |
Xpeng still partially uses maps and radar, while pursuing an AI system centered on vision, much like Tesla. Both aim to develop AI that “drives like a human,” though their philosophies and strategies vary in execution.
5. Where Is Autonomous Driving Headed?
Xpeng’s recent decision reflects a broader industry trend toward vision-based and AI-centric autonomous driving. This evolution is not just about sensor configurations, but a shift in the very concept of a car—from a mechanical device to an intelligent system.
Rather than hardware, the focus is now on data and software.
Rather than precision measurement, it’s about learning and decision-making capabilities.
Autonomous driving competitiveness will increasingly be defined by these elements.
While Xpeng’s transition hasn’t yet reached Tesla’s scale or refinement,
it signals a major shift in China’s autonomous driving landscape,
and positions Xpeng as a noteworthy alternative to Tesla in the global market.
Comments
Post a Comment