Shield AI runs ARMD with Hivemind
Shield AI’s Hivemind autonomy software powered Mitsubishi Heavy Industries’ ARMD drone in recent flight tests, enabling rapid autonomous deployment for Japan’s defense needs. The trial highlights national partners moving experimental autonomy from labs into operational flight testing. (x.com)
Mitsubishi Heavy Industries says mission-autonomy development for the ARMD program began in September 2025 and the full cycle from AI development to UAV installation and flight was completed in eight weeks. (mhi.com) MHI records show the live flight demonstrations took place on November 7, 2025 at a test field in Inashiki District, Ibaraki Prefecture, and on December 18, 2025 at a test field in Ota City, Gunma Prefecture. (mhi.com) Shield AI reported two different ARMD test aircraft performed the live demonstrations, autonomously executing reinforcement-learning behaviors, preprogrammed mission behaviors, and coordinated maneuvers while tracking a virtual air vehicle. (shield.ai) In Shield AI’s account, Hivemind used reinforcement learning to direct the first ARMD to pursue a virtual object and then displayed more aggressive reinforcement-learning motions on the second ARMD during a separate flight. (shield.ai) The teams reported compressing the verification pipeline—advancing from software-in-the-loop (SIL) to hardware-in-the-loop (HIL) testing in under two weeks, then from HIL to vehicle-in-the-loop (VIL) testing in an additional two weeks. (shield.ai) MHI stated that prior in‑house workflows relied on a mix of open-source tools for coding, AI training, simulation evaluation, and HIL testing, and that using Hivemind Enterprise allowed engineers to focus more on mission-autonomy development. (mhi.com) Shield AI’s CTO Nathan Michael described the demonstration as showing Hivemind’s ability to shorten integration-to-flight timelines and enable rapid deployment of autonomy on operational platforms, and MHI said it will strengthen collaboration moving forward. ( )
