The car's black box is used to record sensor data that is used to reproduce and determine the root cause of the accident. Autonomous vehicles at the L3 to L5 levels are increasingly demanding black box recording capabilities, and regulators are developing such requirements, often different from country to country. For example, the Department of Motor Vehicles (DMV) requires self-driving cars to save at least the last 30 seconds of sensor data before a crash.
In addition to all other mechanisms required by law, self-driving cars have a separate mechanism: The car must collect and store at least 30 seconds of autonomous technology sensor data before colliding with another vehicle, object, or natural person in autonomous mode. The mechanism should collect autonomous technology sensor data and store it in a read-only format, and the data is stored until the data is downloaded and stored by an external device, and the data is extracted using this mechanism. In the event of a collision, the data shall be kept for three years from the date of the collision.
Figure 1: Data processing structure diagram in a black box recorder
Figure 1 shows the structure of a black box logger with inputs that bring together a continuous stream of data from multiple sensors. For cost and security reasons, some systems compress data first and then encrypt it. Sensor data bandwidth is mainly consumed by image sensors. A self-driving car will contain up to 12 image sensors, with a remote camera with a resolution of up to 8 megapixels and 60 frames per second. The resulting data stream can reach 20GB/s.
For cost-sensitive applications, H.265 compression technology reduces the final number of bits by 50%, reducing overall storage requirements. H.265/HVEC is a lossy compression algorithm that removes some of the data that is less sensitive to the human visual system. However, some artificial intelligence (AI) algorithms may be sensitive to this kind of data distortion, and when the algorithm reproduces the root cause of the accident based on the record, this compression technique may cause the AI algorithm to run distorted. As a result, some systems, especially those used in robo-taxis, tend to avoid using data compression techniques, or use very low compression ratios. Consumer cars tend to be more tolerant of the use of compression, especially lower-level autonomous driving.
A circular buffer is used to record data for some time before an accident (for example, the last 30 seconds before an accident). A cyclic buffer is memory that can be BASED ON DRAM or flash memory, and there must be enough capacity to meet the storage requirements of a buffer of a certain length. For example, to collect data 30 seconds before an accident, assuming an uncompressed sensor data rate of 1GB/s, the cyclic buffer should have a storage capacity of 30GB.
Typically, flash memory is used to implement a cyclic buffer because it is designed to lose no data in the event of a power outage, while DRAM requires a backup power supply to guarantee that the data collected when the main battery is disconnected is not lost. The difficulty associated with flash technology is overall durability. To take an extreme example, a robot taxi that works 24 hours a day and lasts for nearly 5 years will result in nearly 45,000 hours of continuous operation. Assuming the extreme case of 1GB/s continuous sensor data stream, the required durability will be in the 150PB range. With current flash technology, achieving this level of durability is a real challenge, even a bit impractical.
The NVM storage behind the cyclic buffer in Figure 1 provides a long-term storage location for a 30-second snapshot of data related to an accident or potential accident. The system relies on the results of analysis from an accelerometer (G-sensor) and AI sensors to determine when or is likely to occur. These sensors mark when data in the circular buffer should be written to this long-term NVM memory. Such long-term memory devices are typically flash-based and, unlike cyclic buffers, have much lower durability requirements.
Micron System Architects work closely with our Tier 1 customers and OEMs to design system-level solutions such as black boxes to ensure optimal cost, performance, and power consumption for the most demanding applications. They also want to ensure that system integrators must design applications to prevent any possible memory failures from causing damage to data, people, or property.
With 28 years of continuous cultivation of the automotive market and the most comprehensive range of automotive memory products, Micron is a top automotive memory supplier.
About the Author
Gil Goals
Gil Golov is currently Senior Manager of Automotive Systems Architecture and Strategic Marketing, responsible for Micron's autonomous driving system architecture and solutions. Prior to joining Micron, Gil spent 15 years working on various types of research and development. He holds a Bachelor's Degree in Electronic Science from Tel Aviv University and a Master's Degree in Microelectronics from Brunel University, UK. Gil also holds 26 patents.
Focus on industry hotspots and learn about the latest frontiers
Stay tuned to EEWorld e-headlines
http://www.eeworld.com.cn/mp/wap