Japan Tests AI Multi-Drone Autonomy on ARMD UAV in 8 Weeks with U.S. Shield AI Software.

Mitsubishi Heavy Industries has completed flight demonstrations of AI-powered mission autonomy on its Affordable Rapid-prototyping Mitsubishi-Drone initiative, or ARMD, in Japan, integrating Shield AI’s Hivemind Enterprise and taking the software from development to flight in just eight weeks.

The demonstrations, conducted in November and December 2025 across Ibaraki and Gunma prefectures, validated reinforcement-learning autonomy, multi-drone coordination, and dynamic target pursuit. The ARMD test platform, a 20-kilogram fixed-wing UAV, served as a rapid integration testbed, moving from software development to flight in under two months, a pace aligned with Japan’s push for scalable, AI-enabled unmanned forces.

Related News: Japan FY2026 Defense Budget Positions SHIELD Drones as Core Pillar of Southwestern Island Defense.

Mitsubishi Heavy Industries' ARMD drone completed AI-enabled autonomous flight tests in Japan using Shield AI's Hivemind software, demonstrating rapid progress in mission autonomy for future Japanese unmanned combat and reconnaissance operations (Picture source: Mitsubishi Heavy Industries).

According to MHI’s March 17 announcement, mission-autonomy work began in September 2025, with flights conducted on November 7 in Inashiki District, Ibaraki Prefecture, and on December 18 in Ota City, Gunma Prefecture; Shield AI added that the Gunma event included two live demonstrations using two different ARMD aircraft. That schedule is strategically important because Japan is now treating unmanned mass, rapid software iteration, and simultaneous control of multiple platforms as core features of its future defense architecture.

The test article was a small, engine-powered fixed-wing UAV: 2.5 meters in overall length, 2.5 meters in wingspan, and 20 kilograms in takeoff weight. MHI did not disclose payload, endurance, datalink, sensor suite, or weapon fit, so it would be inaccurate to portray this specific aircraft as an armed strike drone; the disclosed value lies instead in the autonomy layer installed on a compact, low-logistics air vehicle that can serve as a rapid testbed for later operational systems.

What the companies did reveal is operationally significant: Shield AI said the two ARMDs autonomously demonstrated reinforcement-learning behaviors, coordinated motions, and virtual-target pursuit, with the second flight showing more aggressive pursuit logic than the first; MHI said the software package was validated through AI training, simulation evaluation, and hardware-in-the-loop testing before installation on the aircraft. This is a different category of capability from a conventional autopilot following waypoints: it is mission autonomy intended to let an unmanned aircraft sense, decide, and maneuver with much less direct operator control.

Operationally, that opens the door to a wider set of tactical uses than the small size of ARMD might suggest. While MHI did not define the future mission set, the demonstrated behaviors point toward roles such as forward reconnaissance, decoying, target tracking, communications extension, or manned-unmanned teaming in which one operator supervises several platforms rather than hand-flies each one; that is exactly the sort of edge autonomy Japan needs if it wants persistent coverage over dispersed islands and maritime approaches without proportionally expanding manpower.

On the question of armament, the most serious analytical point is that no weapons were disclosed at all. Yet that does not reduce the military relevance of the test, because the decisive “weapon” in this case is the autonomy stack itself: once Japan can field trusted mission software on a small airframe, the same core approach can migrate to future small attack UAVs, anti-ship UAVs, modular UAVs, or collaborative combat aircraft, all of which appear in Japan’s SHIELD concept for inexpensive, large-quantity unmanned forces operating across air, surface, and undersea domains.

This is why the demonstration matters so much for Japan. MHI stated explicitly that mission autonomy is a critical technology for Japan’s UAV operations and that domestic production is essential; that reflects a hard strategic reality, because autonomy software is where tactical logic, mission data, upgrade authority, and long-term sovereignty increasingly reside. A country that does not control that layer may assemble drones, but it does not fully own how they fight, adapt, or evolve in wartime.

The speed of the program is also part of the message: Japan’s defense strategy calls for stronger unmanned defense capabilities, AI-enabled simultaneous control of multiple unmanned assets, and automation to offset a shrinking and difficult-to-recruit force, while FY2026 budget documents describe an urgent need for an asymmetrical architecture built around inexpensive, numerous UAVs, USVs, and UUVs, backed by roughly ¥128.7 billion in requested funding for SHIELD-related unmanned initiatives. In that context, compressing autonomy development, integration, and flight testing into eight weeks is not a publicity metric; it is a readiness metric.

The test also fits a broader Japanese industrial pattern rather than standing alone. MHI has already flight-tested a hybrid medium-sized UAV designed around a 200-kilogram payload objective and a planned 200-kilometer cruising range, while Japan’s defense ecosystem has simultaneously explored AI-enabled combat support drones and loyal-wingman operations; Army Recognition has followed that trajectory in our coverage of Mitsubishi Heavy Industries’ AI-enabled combat support drone concepts, our report on Japan’s preparation for loyal-wingman drone operations, and our analysis of Japan’s mobile laser counter-UAS program.

For defense planners, the most important outcome is not simply that a 20-kilogram demonstrator flew autonomously. It is that Japan and one of its premier defense primes showed they can move from software development to real aircraft testing in under two months, using a domestically prioritized autonomy effort aligned with national strategy; in an era defined by distributed sensing, attritable mass, and machine-speed adaptation, that is exactly the kind of capability foundation Japan must build now if it intends to deter more effectively later.

Written by Evan Lerouvillois, Defense Analyst.

Evan studied International Relations, and quickly specialized in defense and security. He is particularly interested in the influence of the defense sector on global geopolitics, and analyzes how technological innovations in defense, arms export contracts, and military strategies influence the international geopolitical scene.