The development of the Next-Generation Combat Vehicle (NGCV) will require technological advancements in many areas, including lethality, protection, autonomy, human–agent teaming, and electromagnetic capabilities. What ties all of these future capabilities together is the need for vast computational resources to support the artificial intelligence (AI) implicit in bringing these advancements to the battlefield. The operating environment of the NGCV will be such that communications will be severely limited, if available at all; systems will be under constant cyber-attack; and adversarial AI may be actively attempting to deceive all sensors — all occurring under severe size, weight, power, and time-available constraints. These factors, and more, are the motivation for developing a strategy of mobile High-Performance Computing (HPC) for the NGCV.
Future military vehicles will require capabilities beyond autonomous maneuverability, including intelligence analytics and situational understanding, to achieve autonomous operation. Military vehicles must maneuver over and around obstructions, predict and react to various soil conditions, and operate with and adapt to damage well beyond the demands of commercial vehicles. With ever-increasing computational and communications resources placed on vehicles, there is attendant heat generation/ rejection and radio frequency (RF) emission. Signature management, at least for infrared and RF, needs to be a consideration from the outset, rather than an issue to be resolved after the fact.
Maneuverability on the battlefield has many distinct challenges not seen in the commercial vehicle space, including the potential for malicious concealment and deception by adversaries. Military vehicles’ onboard computing will require accurate perception and real-time understanding in a contested and deceptive environment, enabled by robust machine learning algorithms. The path to the NGCV will likely see a proliferation of onboard sensors that generate data that must be processed at the point of need. Examples of these include ultrawideband radar that can scan below the surface of the soil for buried objects, spectrum analyzers searching for adversarial activity, multimodal communications that must be intelligently managed, and acoustic sensors that can detect movement at great distances.
Autonomous vehicles that adapt to and function with damage are critical to robust manned–unmanned teaming (MUM-T) and autonomous vehicles. The U.S. Army Research Laboratory (ARL) recognizes the importance of imbuing teams of heterogeneous robots and sensors with the intelligence to learn and adapt to different settings and perform new tasks along with humans. To achieve the goal of adaptive and resilient teams of robots and humans, significant computing resources must be available onboard the autonomous systems or reachable within a contested and congested environment. The computing resources will support the fusing of sensor data, damage detection and failure prediction/inference, and eventual modifying of operating variables (speed, direction, etc.) to reduce the use of damaged/failing parts.
Training within the vehicle through the use of mobile HPC capabilities will provide an embedded training environment wherever the vehicle is deployed. These training capabilities require the ability to generate synthetic sensor data and drive displays in real time (i.e., emulation with hardware and human-in-the-loop). An example of the screens available in future combat vehicles is seen in Figure 1, alongside computer-generated optical and lidar data.
The automotive industry is moving ahead rapidly with increasing the number and capabilities of integrated circuits (ICs) on vehicles. The automotive growth is driven by systems that provide partial or high automation and that may eventually lead to fully autonomous vehicles; it is boosting total IC content per automobile.
While commercial communications benefit from robust infrastructure with highly engineered cellular network providers interconnected with very-high-bandwidth backhaul links, military operations typically must bring their own communications capability. The military’s mobile devices have very limited access to cloud-based computational capability on the battlefield. Therefore, tactical units must deploy localized edge processing that does not rely heavily upon communications infrastructure. HPC at the tactical edge provides the computational capability at the source of tactical data, the sensors, and users on the battlefield. This array of networked entities constitutes a tactical Internet of Things, termed the Internet of Battlefield Things. Power-efficient edge computing enables the use of machine learning algorithms locally that are integrated with programmable network controllers to intelligently push data over disconnected, intermittent, low-bandwidth networks while minimizing the RF emissions.
Edge computing is enabled through cloudlets, also referred to as micro-clouds, and are localized, trusted, resource-rich computers or a cluster of computers, well-connected to the tactical Internet within one wireless hop — proximity is the key. Cloudlets, just as clouds, are enabled by virtualization. Clouds virtualize an entire computer system using virtual machines, requiring substantial resources. Cloudlets demand a lighter-weight solution, and one option is containers. Instead of virtualizing an entire computer, containers virtualize only the operating system and take advantage of the host computer, such as the Linux kernel, network, and various services. Containers can be tailored to single solutions, such as a machine learning container and a video processing container.
Current tactical radios and electronic warfare systems are packaged as separate point-solutions requiring their own packaging, cooling, processing elements, and antennas. Emerging initiatives seek to establish a common communication infrastructure and processing architecture to consolidate and develop these functions. A significant tactical overmatch may be achieved by fully analyzing electromagnetic (EM) spectral information and combining it with coordinated software-defined radio communications; however, these applications require tremendous amounts of computational power to process the EM signals and execute the associated algorithms. Capabilities such as jamming, direction-finding, spoofing, and stealth communications can all be enhanced and made more efficient with HPC technologies; these may prove to be crucial advantages in future conflicts.
Increased reliance upon software-intensive designs and networked communications increases cyber vulnerabilities in addition to electromagnetic warfare threats. Designing in cyber protection requires cyber testing over the systems’ lifecycle, including vulnerabilities created from integrating multiple disparate systems. In-lab testing of cyber and electronic warfare vulnerabilities through emulation with hardware-in-the-loop (HWIL) is a proven method for evaluation and analysis of integrated systems as part of an LVC (Live/Virtual/Constructive) strategy. HPC on the vehicle will support in-situ emulation, advanced intrusion detection systems, anomaly detection, and machine learning methods to rapidly identify unexpected behaviors.
Integrating all of the future capabilities and intelligence for the NGCV will require a strategy for mobile HPC. Additionally, the challenges and solution look different than the solution for commercial vehicles. The computer architectures will need to support autonomous or assisted mobility, much like commercial vehicles. Additional functionality — such as unrestricted mobility, cyber analysis, and course-of-action analysis — will require a recipe of heterogeneous open-source and commercial off-the-shelf devices. More functionality must be included in the vehicle to support fully autonomous maneuvering in a multidomain battlespace. Figure 2 illustrates this vision and connectivity for mobile HPC that is general-purpose and able to interact and integrate with dedicated computing resources from multiple sources and vendors.
Achieving this vision of advanced AI on the NGCV will require a risk reduction effort that rapidly provides evaluations of technologies and capabilities for transition from basic and applied research to prototypes and live demonstrations. Evaluation of algorithms, software, and hardware capabilities will require hardware and human-in-the-loop testbed capabilities to create a synthetic environment for sensor data, vehicle physics, and so on.
This article was written by Brian J. Henz and Dale Shires of the Army Research Laboratory Computational and Information Sciences Directorate, Aberdeen Proving Ground, MD; Leonard Elliot of the Army Tank Automotive Research, Development and Engineering Center (TARDEC), Detroit Arsenal, MI; and Michael Barton of Parsons Corporation, Columbia, MD. For more information, visit here .