Defense Autonomy: When SDV Enters the Battlefield (Part 1)

The software-defined revolution that started with Tesla is now expanding into the defense sector, and vehicles and weapon systems are becoming “living systems” that evolve based on OTA, fleet learning, and simulation. Applied Intuition is transplanting the Commercial autonomous-driving development ecosystem into defense, enabling large-scale autonomous weapon and fleet operations through Axion (autonomy lifecycle toolchain) and Acuity (AI tactical autonomy stack). The goal is not simple automation, but trusted defense autonomy that continuously improves in the field and can be willingly entrusted by humans - a new battlefield paradigm where humans and AI operate together. AEM met Jason Brown, Head of Defense at Applied Intuition.

Written by | Sang Min Han _ han@autoelectronics.co.kr
한글로보기




Software, after changing the automobile, is now rewriting the battlefield.
Autonomous driving and Tesla’s software-defined concept, which began on roads ten years ago, transformed cars from completed mechanical products into “living systems,” and now that transformation is moving to the battlefield. Code and data determine the fate of the vehicle, functions evolve overnight, and driving data returns as algorithms again. This cycle, modeling - simulation - data collection - validation - OTA - fleet learning - is now moving to battlefields on land, in the air, and at sea.
Applied Intuition - a vehicle intelligence company accelerating global adoption of safe AI-based machines - has, since its founding in 2017, provided OS, SDS, and development toolchains enabling customers to quickly develop and launch intelligent vehicles, serving 18 of the world’s top 20 OEMs. And its solutions have been adopted in major U.S. Department of Defense programs.
In late October, Jason M. Brown, Head of Defense at Applied Intuition, and Dan “Animal” Javorsek, who led DARPA’s ACE program and is now CTO of EpiSci (integrated into Applied Intuition), each shared - in separate settings - their aligned perspectives on the extension of software-defined concepts from automobiles into defense and safe AI.
Brown emphasized OTA on the battlefield and the systems of data and processes required for AI to learn and improve quickly and correctly, while Javorsek, with the question “Is this AI something a warfighter would willingly entrust?” located the essence of defense AI in trust.
Continuous learning, evolution in the field, and systems humans trust and are willing to use - this is what Applied Intuition defines as “defense autonomy.”


 
On October 28 at the Air Force Hotel, during the “International Seminar on the Development of Korean MUM-T Based on Reliable AI,” and in an interview the day before, we spoke with Jason Brown about “how software is transforming the competitive landscape of the battlefield.” (Brown is a 26-year U.S. Air Force veteran deeply connected to Korea. He was stationed at Osan Air Base from 1995 to 1999 and later commanded intelligence organizations. His grandfather was a Korean War veteran; thus, the defense of Korea is a personal issue for him.)

“During my service, and even after taking off the uniform, I have witnessed remarkable advances in defense technology. The change we are experiencing now is a software-defined revolution. Through the Russia - Ukraine war, we realized how much software influences victory in war, and that made me pay great attention to defense software capability. So where did this software-defined revolution start? Tesla. Applied Intuition’s defense business also began with the automotive domain - adding software to GM and ground tactical vehicles,” Brown said.

Tesla and Autopilot. The revolution in the automotive industry where vehicles improve functionality and performance even after leaving the factory. Overnight software updates turned what were once purely mechanical products into living systems that evolve - and now the defense community is paying attention.

“Applied Intuition started from the idea of software-defined systems. Because it has been proven that the source of competitive advantage is not hardware but software and data. And this innovation does not stop at in-vehicle software - it applies to entire ecosystems where large fleets learn and are updated at scale. That is what changed conventional vehicle development.”

Previously, OEMs purchased hardware and software together as black-box packages from suppliers, requiring significant effort to integrate each issue. Adding features or fixing bugs once every few months required massive rework. Software updates were infrequent, expensive, risky, and OEMs had almost no control.
In the software-defined model, hardware and software decision-making are separated. OEMs can access a single code repository that controls the entire vehicle software stack, enabling updates several times a day. Updates are made quickly, tested in simulation in advance, and deployed overnight to millions of vehicles. Time and cost have dramatically decreased.

“The most important point is that OEMs no longer depend on external vendor capability - they own their software destiny. Defense must learn this.”


 
Separation of hardware and software, integrated learning loops, continuous updates - this is the operating principle of autonomous weapon systems.

Brown said, “Ships, aircraft, ground combat vehicles- their hardware lasts decades. Thus, competitive advantage comes from software that evolves inside them. Commercial autonomous driving has already demonstrated this. The ‘autonomy lifecycle’ that Applied Intuition emphasizes is exactly that.”

Over ten years, Commercial autonomous driving developed a mature autonomy ecosystem. This begins with a toolchain that enables the autonomy lifecycle - and that is Axion, the tool system Applied Intuition converted for defense. It includes virtual-environment simulation, high-fidelity simulators for millions of mission miles per day, data management and labeling pipelines for LiDAR, radar, camera sensors, CI/CD for OTA, test and evaluation, and field monitoring systems that feed deployed vehicle data back into the system to retrain models and ensure reliability in a feedback loop.

“Autonomy must continuously be validated across edge cases, and pipelines embed test and evaluation into development. These tools together create the continuous learning feedback loop that defines autonomy.”

Meanwhile, Acuity is the AI autonomy/AI combat software stack created by Applied Intuition using technology from EpiSci, a specialist in defense, aerospace, and uncrewed systems. It enables machines to perceive the real world → decide/plan → control → communicate and maneuver together.

“In the Commercial domain, this stack generally runs inside a single vehicle. But in defense, it must operate across fleets, across domains, and across command structures. That is the biggest difference between Commercial and defense autonomy. Defense autonomy faces an entirely different set of challenges. That is what we considered when designing Axion and Acuity,” Brown explained.

For example, when comparing Commercial automotive and defense autonomy, the two environments are extremely different in data, infrastructure, and operational constraints. Commercial autonomous-driving development takes place in massive datacenters connected by fiber, supported by thousands of engineers. Defense autonomy must operate offline at the edge, without developers.
Commercial companies produce and process petabytes of data; defense often handles gigabytes or terabytes of data from a single aircraft or vessel, processed sometimes on a laptop or secure facility. Commercial systems use dozens of sensors; defense platforms - especially expendable ones - have few, so the challenge is not just developing autonomy but doing so with limited and irregular data. Model testing and deployment also differ significantly.

“Commercial autonomous driving operates under rigorous metrics such as 99.9% accuracy and billions of edge cases, but in defense, 60% accuracy can be enough if it helps operators make faster and better decisions. We are not trying to make a car stay perfectly in its lane - we are helping drones, ships, and robots break through dynamic battlefields under imperfect information. Defense prioritizes robustness and adaptability over perfection.”

Commercial autonomy is always online and cloud-connected; defense operates offline at the tactical edge, in GPS-denied and contested networks. Thus, software must operate without cloud connectivity and think locally. Finally, the users are different - Commercial autonomy supports office engineers; defense autonomy supports analysts, operators, and warfighters.
“They do not want dashboards full of data - they want clear and trustworthy real-time outputs. Commercial AI laid the foundation for autonomy, but defense adapts it for environments where connectivity is uncertain and human lives depend on trust and clarity. Such systems must be rugged, modular, and self-sufficient. They must be software that improves even in the smoke of battle. Thus, the defense version of the autonomy lifecycle must combine Silicon Valley agility with battlefield resilience. And we must look at what enables large-scale autonomy.”






 
“Armed forces worldwide, including the U.S. and Korea, still tend to operate small-scale exquisite systems. Each platform is custom-built, extensively tested, and manually integrated. That model worked for decades. But in cooperative systems - fleets of aircraft, ships, and vehicles - a completely different approach is needed. We need software that can be deployed, tested, and improved at scale,” Brown said.

Commercial autonomy already solved this problem. To build trustworthy autonomy,Commercial programs perform tens of millions of simulation miles daily. Simulation safely explores rare and dangerous edge cases, then real-world data validates them, creating a loop that builds trust before deployment. The system evolves with the platform, not waiting for new hardware.

“Tesla and Waymo evolved from basic lane-keeping to full autonomy through iteration. It is all about data - the more data, the faster the learning. With each software update, the system becomes smarter. The more data collected, the faster the improvement,” Brown said.

Defense autonomy is the same. Instead of a few exquisite prototypes, what matters is a continuously improving fleet - unmanned surface vessels, drones, and ground robots operating together, learning from each mission, updating nightly, and sharing lessons across the force. Defense must move from demonstration to superiority through this method.

“Allies must build software ecosystems that enable true collaborative autonomy. Large-scale autonomy is not just a technical concept. It is a new defense operating model - treating data and software as the decisive weapons of the 21st century.”


 
What is the true value of software-defined systems? It changes not only how systems are built but who benefits - both system integrators (OEMs and defense organizations) and end-users (soldiers).
First, cost reduction and simplification. Centralized computing and hardware/software separation can reduce hardware complexity and integration cost by up to 10×. Instead of redesigning physical components, new software can be pushed to existing hardware - faster, cheaper, easier to maintain.
Second, modularity. A software-defined architecture abstracts hardware specifics. Sensors, processors, radios can be replaced without rewriting entire code. In defense, this is crucial - rapidly upgrading sensors or payloads across fleets without recertifying entire systems.
Third, agility. In software-defined vehicles, deep-layer updates (targeting algorithms, navigation models, communication modes) are deployed to the field within days. This provides capabilities adversaries cannot quickly follow and accelerates battlefield evolution.
Fourth, performance. Centralized software architectures optimize compute allocation across platforms, delivering reliable performance and better management even under power and bandwidth constraints.
Fifth, trust. The value of AI is only as high as the trust operators give. Continuous updates deliver safer and more reliable models to the battlefield without waiting for new hardware cycles. Because operators directly observe improvement, trust builds over time.
Lastly, MUM-T (Manned - Unmanned Teaming).

“Software-defined systems allow repeated testing of complex human-machine interactions in simulation, labs, and real environments - without disrupting real missions. In this way, we achieve genuine collaboration between humans and machines, not mere interoperability.”



 

AEM(오토모티브일렉트로닉스매거진)

---