Accelerating the Design and Deployment of Autonomous Cars Worldwide

November 10, 2020


Accelerating the Design and Deployment of Autonomous Cars Worldwide

To meet market demands, designers of autonomous cars will need to rely heavily on off-the-shelf technology. There simply isn’t the time to for OEMs to design everything in-house.

Autonomous cars are an important technology globally. Vehicles that can drive themselves will improve automotive safety by eliminating major causes of accidents, including driver distraction and drowsiness. Smart vehicles will also be able to address traffic issues by safely operating more cars on existing roads through coordinated driving to mitigate traffic. This will be achieved by enabling cars to follow each other more closely than is safe for independently controlled vehicles relying on human reaction speeds.

There is tremendous momentum to place autonomous vehicles on the road as soon as possible. From a technological standpoint, the necessary hardware is starting to be in place to provide sufficient processing power for the many layers of real-time artificial intelligence that must be located in-vehicle. Significant progress is also being made in terms of the software required to drive these vehicles in a safe and reliable manner.

At the same time, however, there is increasing pressure to accelerate the development of autonomous vehicles. Autonomous cars will offer much more than increased efficiency and safety. They will also help countries meet pending workforce shortfalls that could have wide-ranging negative impact on businesses and urban development if not addressed.

The Pressing Need for Autonomous Vehicles

Consider the importance of cars for personal and commercial transport in Japan. The number of newborns each year in Japan continues to drop, leading to a decrease in the available workforce. Combined with the fact that young people are choosing careers other than driving, this has led to a diminishing workforce with fewer new drivers to replace elderly drivers as they retire.

At the same time, the shift to a modern convenience lifestyle – 24-hour couriers, delivery services, logistics, and mass transportation – is increasing the demand for drivers. As the elderly population continues to rise, this will only put more demand on these types of services.

The current workload is already too much for the existing workforce, and they are unable to meet even today’s demands. Solutions like overnight shifts and increased overtime will bolster the available driver pool. However, they do so at the tradeoff of straining already overworked drivers. The driving conditions throughout Japan are considered to be challenging and dangerous, and this added strain will only lead to more traffic accidents.

The driver shortfall has already begun to impact the nation by forcing some businesses and services to slow or cease operations. It is clear that autonomous cars are particularly important for Japan’s future.

However, these trends and cultural factors are not unique to Japan. They impact every country and every major city in the world. These factors are also exacerbated by the COVID-19 pandemic. While individuals are driving less, they are purchasing more goods and services online, adding further stress and demands on the diminishing driver workforce. Autonomous vehicles will play an essential role in the evolution of infrastructure in modern cities. The challenge is how we get there as quickly as possible.

Overcoming the Technological Challenges of Autonomous Cars

To meet market demands, designers of autonomous cars will need to rely heavily on off-the-shelf technology. There simply isn’t the time to for OEMs to design everything in-house. Increasing time-to-market pressures will also require OEMs to focus on technology that is ready and available today.

For example, the ideal autonomous car is an electric vehicle (EV) that will also reduce pollution, another important problem for highly populated areas. Unfortunately, EV technology is still in its infancy. In addition, the infrastructure required to support cities of EVs are still years away from being in place. Thus, the first wave of autonomous cars will be built using traditional combustion engines and other technology that already exists today.

Complicating the challenge is the limited space within the vehicle to house computing equipment and cables. To address space limitations, solutions will have to be highly integrated. In addition, the harsh operating environment of a car – including high temperatures, vibration, and rough movement – is nothing like the controlled conditions of a computer lab or enterprise network server room. Hardware will need to be able to operate reliably at automotive temperatures.

Performance limitations are also a key consideration. To become autonomous, each vehicle requires a computer with sufficient performance to be able to handle real-time data transfer of video to an on-board artificial intelligence (AI) engine with the ability to process and control the vehicle in real-time as well. In addition, to be able to minimize footprint, maintain real-time responsiveness, and avoid signal incompatibility issues, all computing and video equipment must be able to integrate into the vehicle’s system infrastructure. This means being able to communicate with and control the car using a CAN interface. Another complication is that 5G networks are not ready yet, so high-speed vehicle-to-control center and vehicle-to-vehicle communications will be constrained.

Sufficient storage is another concern. There are many capacity issues to consider, including having enough local storage to perform AI computing while supporting real-time video backup.

Finally, all of these systems must be able to operate with high reliability while running off the “dirty” power provided by any battery connected to an engine.

Each of these issues must be taken into account early in the design process to ensure high reliability. This is critical in order for OEMs to make the shift from SAE Level 3 (conditional automation with a human driver) to SAE Level 4 (high automation without a human driver).

Integration for Reliability and to Accelerate Design

To address these challenges and get autonomous vehicles on the road quickly, a high-level, integrated approach to computing is required. Rather than building individual subsystems that have to interact together, a high-level approach focuses on a workstation-grade platform that is purpose-built to meet the processing requirements of today’s vehicles with Advanced Driver Assistance System (ADAS) functionality. From this integrated foundation, more advanced autonomous capabilities and EV technology can be introduced as they become more readily available.

Taking a high-level approach has several key advantages. First and foremost, it builds off of existing vehicle technology and enables OEMs to significantly accelerate their design cycle. In addition, because the workstation-grade platform is pre-integrated, it become possible to use best-in-class technologies without compromising reliability or delaying time-to-market. Since the platform is purpose-built for automotive applications, it can provide the processing resources necessary for vehicular AI computing, on the order of 2300+ CUDA cores.

Using a platform also allows for a more compact solution. An integrated system reduces the amount of cabling required and the need to run cables through the vehicle where there is little space available. From an interface standpoint, with a CAN interface, a platform can consolidate communications through a single point of entry for the entire autonomous computing engine, simplifying communications and minimizing the impact on the vehicle’s infrastructure. This, in turn, reduces the overall cost of installing an autonomous system inside a vehicle.

To provide sufficient storage capacity in a compact size as well, Solid State Drives (SSD) are available, which are more resilient to vibration and movement than traditional hard disk drives (HDD). SSDs are built for rugged applications and are lockable and hot-swappable for mission-critical data protection. They can further increase reliability through RAID data protection technology.

Consolidating all computing resources in a single platform also simplifies power management. Ignition control is essential for computer equipment operating off a car battery, as is support for a wide range power input with surge protection. A platform can “clean up” power, filtering out engine noise and other disruptions, to provide reliable and consistent power to all sensitive electronics essential for reliable operation.

The result of all these benefits of taking a platform approach is faster time-to-market and lower project cost, both important factors to help accelerate the deployment of reliable autonomous vehicles. A platform also gives OEMs the flexibility to customize the resources available to the AI system. For example, a platform can house additional memory drivers and/or higher performance computing boards/AI accelerators to match the requirements of the particular application.

The ECX-1400 PEG AI Computing + NVIDIA GeForce RTX 2070 Graphics Card

Vecow has developed a purpose-built platform for autonomous vehicles. The ECX-1400 PEG AI Computing + NVIDIA GeForce RTX 2070 graphics card is a compact, full AI system solution that provides workstation-grade AI performance with an eight-core Intel Coffee Lake processor and Intel C246 backed by NVIDIA’s revolutionary Computer Unified Device Architecture (CUDA) core technology (see Figure 1). The NVIDIA RTX and Turing GPU architectures provide up to 6X performance for AI computing, at up to 8K resolution. The fanless system was designed to support an independent graphics card to provide maximum flexibility while keeping a compact form factor.

Figure 1. The ECX-1400 PEG AI Computing + NVIDIA GeForce RTX 2070 graphics card is a compact, full AI system solution that provides workstation-grade AI performance with an eight-core 9th Gen Intel Coffee Lake Refresh processor and Intel C246 backed by NVIDIA’s revolutionary Computer Unified Device Architecture (CUDA) core technology.

The platform has 4 lockable and hot-swappable 2.5 inch SSD trays to provide up to 32 TB of reliable storage for mission-critical data protected with RAID 0, 1, 5, or 10. It also supports the high-speed communications required for autonomous car applications, including up to 7 independent HD displays (VGA, HDMI, DVI, DP), USB 3.1 Gen 2 (10 G), PCIe 3.0 (8 GT/s), SATA1 III (6G), and USB 3.0 (5G). The availability of 4 PoE+ interfaces reduces cabling requirements, and a customized mechanical design supports CAN bus ports in current vehicle systems.

From a power perspective, the ECX-1400 PEG AI Computing platform is well-protected. With an optimized power design capable of providing a 300 W power budget for the graphics card, the platform has 32 isolated device input/output (DIO), ignition power control, smart circuit protection, 80 V surge protection, and 12 V to 36 V DC power input. Smart power management capabilities include ignition mode management, iAMT 12.0, TPM 2.0, PXE, Wake on LAN, and remote power switch. The system is also designed with anti-shock and anti-vibration protection and can operate from -20 °C to 45 °C.

Autonomous Cars in Action

The ECX-1400 PEG AI Computing + NVIDIA GeForce RTX 2070 graphics card has undergone numerous field trials. Two cases are shared here. The first is for a convoy of autonomous trucks (see Figure 2) driving goods from a central warehouse to various destinations in Japan.

Figure 2. Autonomous trucks delivering goods to various destinations.

Controllers placed in the lead and rear trucks manage high-level tasks like fleet communication, speed/distance control, route planning, sensor identifying, ambient surveillance, and fleet video/data sharing for the entire convoy (see Figure 3).

Figure 3. With the ECX-1400, two controllers can manage a convoy of autonomous trucks from warehouses to destinations across Japan.

The second case is an autonomous shuttle bus for transporting passengers from the airport to various stations around Japan (see Figure 4).

Figure 4. Autonomous shuttle buses transport passengers across Japan.

The bus is managed remotely from a dispatch center, and its sensor system utilizes magnetic markers on the road for navigation (see Figure 5).

Figure 5. With a platform like the ECX-1400, autonomous shuttle busses can transport individuals from the airport to stations around Japan.

This video shows how a flexible real-time computing platform can proving the AI processing capabilities necessary to enable autonomous vehicles today.

The ECX-1400 PEG AI Computing + NVIDIA GeForce RTX 2070 graphics card is a highly flexible platform and can be customized for specific applications. Although purpose-built for autonomous vehicles, it can be used for other applications including automatic license plate (LPR) recognition, robotic control, telemedicine, and public surveillance.

With the capabilities of the ECX-1400 as a single platform solution, it is possible to overcome many of the challenges associated with integrating autonomous car AI equipment into vehicles. This technology simplifies the design of autonomous cars and will accelerate their deployment throughout the world to help make our roads safer and address increasing driver workforce shortages.