The development of autonomous driving technology can be divided into five stages:
-
1-hour autonomous driving: Achieving basic functions, capable of autonomous driving for about 1 hour. The key is vehicle modification and basic capabilities.
-
10-hour autonomous driving: Mainly relies on the progress of various machine learning models.
-
100-hour autonomous driving: Requires large-scale data collection and complex model training. The key is to establish a complete data collection and simulation training system.
-
1000-hour autonomous driving: The core is to establish a scientific evaluation metric system that can accurately assess system performance improvements.
-
10000-hour autonomous driving: Needs to consider overall traffic safety, not just self-safety, but also reducing risks to other vehicles. The system has surpassed human level and needs to establish self-learning and evolution mechanisms.
In this process, key points include:
- Evolution from basic functions to complex models
- Collection and utilization of large-scale raw data
- Establishing a scientific evaluation metric system
- Self-learning ability after surpassing human level
- Considering overall traffic safety, not just self-safety
The progress of autonomous driving technology is a long process, with each stage taking 1-3 years. Currently, industry leaders have reached over 1000 hours of autonomous driving and are moving towards 10000 hours.
Views on data:
- When the system surpasses human level, human driving data may become a "disturbance"
- Need to filter high-quality data rather than simply pursuing data quantity
- Establishing self-learning and evolution mechanisms is more important than simply inputting data
Overall, autonomous driving technology is shifting from "resource-driven" to "capability-driven", with evaluation systems and self-evolution capabilities becoming key factors.