Systems to Record Raw ADAS Sensor Data During Test Drives
Organizations that are developing ADAS (advanced driver assistance systems) require test vehicles to collect and store raw field camera and sensor data. The collected data is used to train the software that is used by the vehicle computer to enable ADAS features, such as lane keep assist, adaptive cruise control, and autonomous self-driving capabilities.
Furthermore, the data collected by test vehicles (camera footage, sonar, radar, GPS, and Lidar) is often used to train vision and object detection algorithms. Data and a lot of it are required to train the computer vision and machine learning models that are used to make advanced driver assistance systems possible.
The sensor and cameras that are added to vehicles to enable ADAS generate a mind-boggling amount of data. Intel estimates that autonomous vehicles equipped with Cameras, Lidar, Radar, and GPS sensors can generate 4TB of data per vehicle per day. So, for organizations that want to record and store sensor data to train ADAS software, they must equip vehicles with powerful systems to record raw ADAS sensor data.
Computer system used to record RAW ADAS sensors data must have the necessary processing power to gather data from the sensors and cameras. Once the data is gathered, edge computers must be configured with a large amount of high-speed data storage to store the Terabytes of data generated by vehicle cameras and sensors.
ADAS recording computers are hardened to withstand deployment in challenging environments that are not suitable for regular, consumer-grade desktop PCs. Furthermore, industrial edge computers are equipped with rich I/O enabling systems to interface with various devices thanks to the inclusion of Ethernet ports, Serial COM ports, USB Type-A ports, and GPIO (general purpose I/O). Additionally, systems can be configured with Terabytes of SSD (solid-state drive) storage or HDD (hard drive) storage. Having plenty of storage is a must when recording raw high-resolution camera and sensor data, especially when you consider that ADAS can generate 4TB to 5TB of data per vehicle per day.
ADAS Data Recording
ADAS data recording computers must be powerful enough to record a continuous and uncompressed flow of data from cameras, Lidar, radar, GPS, and vehicle buses. Every little bit of data must be collected so it can later be used to train machine learning and machine vision algorithms for ADAS and autonomous driving (AD). ADAS and autonomous vehicle computers must be equipped with plenty of performance to process, analyze, and diagnose sensor data provided by test vehicles on the road most of the time, recording the surrounding environment to train ADAS and autonomous vehicle algorithms.
Recording data for ADAS is not an easy task. Typically, organizations have large fleets of test vehicles that drive millions of miles per year, recording the environment and surrounding conditions in order to cover all of the possible road conditions and objects that a vehicle may encounter. The more data and the better the data collected, the better the trained model will perform when it encounters new environments and objects that it has never seen before.
Moreover, test vehicles often go to icy environments and environments that are extremely hot, so the computing solutions that are onboard the vehicles must be robust enough to cope with the extreme heat and extreme cold they’ll encounter. As such, edge computing solutions are configured with a wide operating temperature range, ranging from -25⁰C to 60⁰C. Whether a vehicle travels to the Mojave Desert where the temperature reaches 50⁰C or New York during the winter where the temperature reaches -15⁰C, the systems are hardened enough to continuously and reliably record raw sensor data avoiding data loss and corruption.
Additionally, ADAS and autonomous vehicle data capture computers are often placed in the trunk of a vehicle, so they must also be compact enough to fit in a vehicle’s trunk while operating reliably to capture all of the sensor data. Moreover, computers must be equipped with high-speed data storage because they must write Gigabytes of data to the drive per second to capture the real world data required to train the machine learning and deep learning models enable the ADAS.
Premio AI Edge Inference Computing Solutions
Premio offers a variety of AI edge inference computing solutions that can be configured with several choices of CPUs, storage solutions, RAM, and I/O ports. This allows the system to continuously capture uncompressed data from high-resolution cameras, Lidar, radar, automotive buses, and other devices that will be used to train machine learning and deep learning models.
Furthermore, Premio’s AI edge PCs are hardened to endure challenging environmental conditions, such as exposure to dust, debris, dirt, shock, vibrations, and extreme temperatures, allowing the system to operate 24/7 without any interruption to data acquisition and recording tasks.
Moreover, AI edge inference computing solutions are designed and built with shock and vibration resistance, enabling them to withstand frequent exposure to shock and vibration without sustaining damage to the sensitive internal components. In fact, Premio’s solutions come with 50Gs of shock resistance and 5GRMs of vibration resistance in compliance with the MIL-STD-810G. This makes them more than capable of handling exposure to shock and vibration in compliance with military standards.
Shock and vibration resistance are essential for computing solutions deployed in vehicles because vehicles constantly move, exposing computers on board to shock and vibration. Regular computers are not designed, nor are they built to withstand deployment in such environments. However, Premio’s solutions are specifically designed and built to endure such challenging environments.
Additionally, Premio’s AI edge computing solutions are passively cooled, meaning that fans have been removed from the system. The removal of fans means that we’ve had to rely on passive cooling via the use of heatsinks to cool down systems. Although active air cooling provides the best cooling, heatsinks offer an efficient way to move heat away from the system’s sensitive internal components to the system’s outer enclosure, which dissipates the heat into the atmosphere.
That said, the elimination of fans has allowed us to design a totally closed system, eliminating the ability of dust and other small particles to enter the system to damage the internal components. Additionally, fans are one of the leading parts that fail in electronics, so by removing them we’ve eliminated a component that commonly fails, making our systems more reliable and durable, reducing the possibility that you’ll experience downtime by deploying AI edge computing solutions.
Having said that, if you choose to configure an AI edge computer with a GPU, the system will be split into two compartments. The main compartment that houses the CPU, RAM, chipset, and storage devices is fanlessly cooled, while the compartment that houses the GPU is actively air-cooled via the use of fans. This is so because GPUs consume a lot of power and therefore produce a lot of heat. Heatsinks are not sufficient to cool down GPUs; therefore, fans must be used to cool them down and expel the heat generated inside the system. That said, while not completely fanless when configured with a GPU, AI edge computing solutions with a GPU are still rugged and can be deployed in many of the environments that totally fanless rugged computers can be deployed.
Moreover, edge AI computers are equipped with a wide power input, making systems compatible with a variety of different power input scenarios thanks to a wide power input ranging from 9 to 50 VDC. Also, systems are configured with a number of power protection features that include overvoltage protection, surge protection, and reverse polarity protection. Solutions that are deployed in vehicles must have a wide power input because they are often powered from vehicle batteries with different voltages.
Furthermore, ADAS edge computing solutions are configured with power ignition management. Power ignition management detects when a vehicle has been turned on, allowing a system to initiate a boot delay. Also, power ignition management senses when a system has been turned off, initiating a delay before shutting down the system. The delay allows the system to finish the tasks at hand, helping organizations avoid the loss or corruption of their data by a sudden system shutdown. Also, power ignition management prevents the system from draining a vehicle’s power while the vehicle is turned off.
Bottom Line
The bottom line is that the more data the makers of ADAS have, the better they can make their advanced driver assistance systems or autonomous driving systems work. The makers of advanced driver assistance systems and autonomous driving systems must collect and store data so that the data can be used at a later time to train algorithms. To do this, organizations need powerful computing solutions that have plenty of storage and can be deployed at the edge to record the raw data generated by high-resolution cameras, sensors, networks, protocols, and vehicle buses. Premio offers a wide variety of configurable AI edge inference computers that can be configurable with specific processors, storage, memory, and even connectivity.
