Ģ����Ƶ

Autonomous platforms will perform a range of functions on the battlefield, from intelligence, surveillance, and reconnaissance (ISR) and close combat support to communications relay. They will be flexible and adaptable: A single unmanned autonomous vehicle (UAV) could undertake a variety of mission sets in subsequent days. They will be “attritable”—designed to prioritize function and expendability rather than durability. In addition, they will operate together as intelligent systems, collaborating tactically with minimal human intervention.

One key advantage of autonomous platforms is they will require lower data transmission rates than remotely piloted vehicles, which will enable them to operate for extended periods in austere, degraded, or denied communications environments. Whereas a remotely controlled drone needs to be able to stream video back to its pilot, an autonomous ISR asset can pilot itself; it only needs to send back the coordinates of a target and the probability of successful identification, for example. Reduced communication requirements relax data latency constraints, message complexity, and exploitability, requiring less power and allowing advanced jam-resistant encoding.

Attritable by design, aircraft sent to swarm enemy lines will have a much shorter life span than today’s crewed platforms, which in turn will translate to a new model for sustainment budget and infrastructure. This shift is already underway: The Air Force retired the original MQ-1 Predator drone less than 15 years from its initial operating capability, while many manned aircraft have been flying for many decades.

Throughout history, adjustments in tactics, techniques, and procedures were disseminated through human channels, from the chain of command to the soldier in the trenches or the pilot in the cockpit. Mission leaders determined combat loads based on immediate operational needs, and an infantryman needed no extra training to carry a load on a multiday dismounted patrol one day and a combined arms engagement the next.

Now, autonomous platforms will draw from a vast library of mission-specific software and algorithms. Just as soldiers can only carry so much on their backs, the size, weight, power, and cost (SWAP-C) constraints on an autonomous platform’s storage and compute capacity limit its combat load to the algorithms, software, and data necessary only for the mission at hand. What this means in practice is that a mission involving a kinetic strike will require a different software payload than a medical evacuation. Other data- and compute-intensive uploads could be systems for language recognition systems (e.g., Cyrillic street signs versus Cantonese) and terrain navigation in GPS-denied environments (e.g., an asset deployed in East Africa won’t need detailed maps of coastal Guangdong Province), or computer vision algorithms for target recognition. The Pentagon will have to develop and continuously improve procedures for pushing updates to the fleet.

SWAP-C limits also require sophisticated machine learning operations (MLOps) to sort intelligence from noise. This is particularly crucial for long-duration missions in communications-denied areas, such as subsurface drones operating near adversary coastlines. These underwater vehicles must operate autonomously for extended periods. They require advanced algorithms to sharpen collected data: to distinguish between irrelevant environmental features (e.g., images of empty sea) and objects of military interest (e.g., ships, submarines, mines). Sharpening the data reduces the amount of storage capacity autonomous vehicles require. Upon completing their mission, these drones can resurface at predetermined locations to rapidly transmit relevant data to secure cloud systems. An array of plug-in hardware modules will enable versatility, such as different wing sets for loitering ISR or ground attack, and different loads for electronic warfare or a kinetic strike. The evolution of versatile autonomous platforms will also encourage innovation in hardware and sensors.

The recent history of warfare offers countless examples of tactical updates that couldn’t have been foreseen before a platform was deployed to the field. Software requires updates to optimize performance, even in relatively static environments. Permissive and collaborative environments for complex systems, such as self-driving cars, can require frequent updates to improve safety and the driving experience. The contested world of the modern battlespace with an adapting adversary increases this need exponentially. From new adversary weapons, sensors, and signatures to refined tactics, optimal behavior will require daily or even hourly deployment of software updates.

The technical, logistical, and bureaucratic challenges to bring autonomy to military operations, and soon, demand an unprecedented integration of hardware, software, algorithms, and human teams, both at the enterprise level and the edge.

Autonomous military operations will consist of multitudes of autonomous platforms and enterprise systems. Updates on one platform will need to be coordinated and synchronized across the system of systems. An autonomous submarine and an autonomous aircraft will have different manufacturers but will need to work together, with one sweeping for mines while the other provides aerial overwatch.

Intelligent systems consisting of multiple autonomous platforms will demand hyperscaler-level data and compute capacity. The logistical infrastructure to move all the data and algorithms must be treated as a critical warfighting system in its own right. The data itself is a strategic asset; in its “,” the Defense Information Systems Agency noted “inherent power in owning data to control the high ground.” The data and communications infrastructure will need a robust cybersecurity wrapper.

For each of the diverse mission scenarios and autonomous platforms undertaking them, developers must create, verify, and validate updates, and the technical and logistical complexity of disseminating them to the fleet securely is immense. A future force will need to update tens of thousands, or even hundreds of thousands, of unmanned systems simultaneously, creating the potential for an immense data bottleneck—envision the difficulties football fans have using mobile devices in the stadium during a game. That effort will demand advanced edge-based routing to move data on the best available network without human intervention.

A combatant command could soon store 10,000 powered-down UAVs in a warehouse for months until an emergent event. The force would need to update and synchronize these crafts as quickly as possible for deployment, including fresh software and algorithms. An autonomous civilian car can take multiple updates per week in a permissive communications environment. Military operations can take place in contested, non-permissive environments where the connectivity required to receive updates is not always a given.

A new paradigm for military operations demands a new paradigm for software development and acquisition. The move toward attritable autonomous systems will bring wholesale change in the economics of national defense, notably in a shift in the Pentagon’s budget from conventional operations and maintenance to software. It will also require DOD to think differently about data. To ensure the continuous improvement of autonomous platforms, it will be imperative to build capabilities that can gather and store as much data as possible in permissive training environments with less network constraint. This data will train the algorithms that will serve as the brains of autonomous vehicles and weapons.

DOD’s models for tech acquisition were developed over decades to suit the purchase of giant, costly pieces of hardware or massive enterprise software. Those acquisition models will need to be nimbler and more flexible to keep up with technological innovation. Under current procurement and operations and maintenance models, a platform’s original contractor is typically tasked with maintaining and updating its hardware and software. But the startup or tech company that manufactured an autonomous aircraft may not be best suited to write mission-specific algorithms; its engineers and executives may not have the necessary experience working with federal clients or the security clearances to fully understand the nature of the operation.

Original equipment manufacturers typically lock in sustainment funding on an exquisite, sophisticated legacy platform for the decades it’s in operation. Attritable platforms with short life spans won’t need as much sustainment, but they will need frequent software updates. The lower barrier to entry into the software market as opposed to developing and manufacturing, say, a crewed fighter jet, will enable more companies to pursue innovations, with greater efficiency and resilience in the supply chain. Furthermore, it may not make financial sense for every private-sector company that makes high-quality drones to invest in customizing its product for the highly specific parameters of defense operations. As autonomy becomes increasingly essential to our nation’s defense, DOD may need to retain control of the algorithms that govern the behavior of autonomous systems. By keeping the code separate and owning it, DOD can buy the best available drones and UAVs without having to rely on manufacturers to tailor the product to the contours of ever-changing missions.

Divorcing the acquisition of hardware from the acquisition of software also will enable DOD to buy the latest and greatest from an expanded Silicon Valley vendor base working in concert with prime defense contractors. Imagine a software package that excels at performing post-mission analysis that is available to every squadron that uses UAVs, regardless of which manufacturer makes the hardware. Autonomous systems will continue to evolve and acquire newfound capabilities that can serve a variety of key mission functions. Pivoting to a software-first mindset will help position DOD to bring the most innovative technologies to bear on the greatest challenges.

• To successfully deploy tactical autonomy in U.S. military operations, the Pentagon will need to rethink how it purchases and deploys one of the foundational elements of technology: software.
• Autonomous platforms will draw from a vast library of mission-specific software and algorithms and will require regular updates based on the missions they are set to perform.
• Divorcing the acquisition of hardware from software also will enable DOD to buy the latest and greatest from an expanded Silicon Valley vendor base working in concert with prime defense contractors.