DarkstarDarkstarBecome a design partner
Swarm autonomy testing

Testing drone swarms and collaborative autonomy

Many uncrewed systems coordinating with one another is one of the hardest things to test, because behavior can emerge from interactions no single vehicle test reveals. This page collects what official sources say about attritable autonomous systems, swarm experiments and the limits of testing them.

Public sources, checked October 14, 2026.

What the public record says

Replicator, unveiled August 28, 2023 and led by the Defense Innovation Unit, aimed to field thousands of uncrewed systems by August 2025. A January 2026 Congressional Research Service report notes one former defense official said only hundreds had been fielded by summer 2025. Source

The Defense Innovation Unit describes its all domain attritable autonomous systems as less expensive, putting fewer people in harm's way, and changeable with much shorter lead times. Source

In November 2024 the Defense Innovation Unit named Swarm Aero, Anduril and L3Harris for Autonomous Collaborative Teaming software meant to coordinate large groups of uncrewed assets across domains, after 132 companies submitted 165 solutions. Source

DARPA's final OFFSET field experiment combined two testbeds totaling more than 300 small commercial air and ground platforms, with virtual swarm agents operating alongside the physical robots. Source

Per the Congressional Research Service, DoD Directive 3000.09 (Autonomy in Weapon Systems, updated January 2023) requires covered autonomous and semi autonomous weapon systems to be tested to confirm they work as intended in realistic operational settings, including against adversaries using realistic countermeasures, and retested after changes. Source

Why testing is hard

  • The Congressional Research Service says swarming, meaning uncrewed systems autonomously coordinating with one another, would likely require further advances in artificial intelligence and networked communications. Source
  • An OSD sponsored 2020 study says that for systems with possible emergent behavior the test design space grows with each added factor, and that continuity of factor interactions cannot be assumed for autonomy. Source
  • The same study calls simulation unavoidable and necessary for validating autonomous systems but only as good as the model behind it, and suggests blending simulation with partially real environments. Source
  • GAO found in December 2025 that DoD wide test policies only partly reflect leading practices such as involving testers early, iterative testing, and digital twins. Source

Where Darkstar fits

Darkstar is an early stage simulator project for replaying adversarial scenarios against autonomous systems on real terrain, designed for drone and swarm autonomy teams first. It is not a finished product: design partners help decide which scenarios it replays first, so the work fits their own evaluations.

Other segments

Guides

Darkstar by Think Tech Holdings

Find autonomy failures in simulation, not in the field

Darkstar is an early stage simulator project for replaying adversarial scenarios against autonomous systems on real terrain. We are looking for design partners to shape what it tests first.

Become a design partner
DarkstarGuidesPrivacy and text terms