*What It Is * FleetLearn is a GitOps-based system for securely distributing and managing AI models across autonomous platforms operating in denied, disrupted, intermittent, and limited (DDIL) environments using tactical mesh networks—without requiring continuous cloud or SATCOM connectivity.

FleetLearn extends Rancher Fleet to support JADC2-style distributed operations where decision authority and intelligence must exist at the tactical edge, not just in centralized data centers.

*Why This Matters for JADC2 * JADC2 requires:

Distributed decision-making

Rapid adaptation across domains

Resilience under adversarial network conditions

Shared, trusted situational awareness across heterogeneous platforms

Today, AI intelligence does not move at machine speed in contested environments.

Most AI deployment pipelines assume:

Centralized cloud orchestration

High-bandwidth backhaul

Stable control planes

At the tactical edge, those assumptions fail.

Without a way to safely update AI models mid-mission, JADC2 systems risk:

Acting on stale intelligence

Diverging behavior across platforms

Fleet-wide failure from a single bad update

Loss of trust in autonomy

FleetLearn addresses this missing JADC2 building block: AI capability distribution under fire.

*Who This Is For * FleetLearn is designed for:

DoD programs deploying autonomous or semi-autonomous platforms

Swarm-based systems (UAS, counter-UAS, ISR, interceptors)

Teams operating tactical mesh radios and edge compute

Platform and autonomy leads responsible for fleet reliability and mission safety

This is not enterprise MLOps. It is operational AI infrastructure.

What FleetLearn Enables (JADC2 Lens)

FleetLearn enables federated learning at the tactical edge, without centralized dependency:

Distributed Intelligence Updates

AI models propagate across the force opportunistically

Command Resilience

Enables operations even when higher echelons are unreachable

Trust and Control

Every update is verified, versioned, and reversible

Cross-Domain Scalability

Works across ground, air, and maritime platforms running Kubernetes at the edge

Each node pulls updates independently, but the force adapts coherently.

What Needs to Be Built (Extending Rancher Fleet)

FleetLearn is not a rewrite—it is an extension of Fleet for tactical reality.

Additions to Fleet

Mesh-Aware Transport Layer

Support for tactical mesh radios (e.g., Silvus, DTC)

Latency- and loss-tolerant sync behavior

Adaptive throttling based on link health

Multi-Hop Propagation Logic

Nodes act as peer distribution caches

Intelligent forwarding based on network topology

Prevents redundant transfers across limited links

AI Model–Optimized Packaging

Chunked and delta-based model updates

Compression and content-aware splitting

Avoids full-image redeploys for minor changes

Tactical Safety Controls

Canary deployment across subsets of platforms

Mission-aware rollout policies

Automatic rollback on degradation or misclassification indicators

Cryptographic Trust and Provenance

Signed model artifacts

Verification at every hop

Auditability aligned with zero-trust principles

Disconnected Operation by Default

No assumption of centralized control plane

No requirement for clock sync or GPS

What the Demo Proves

The demo is designed to mirror real-world JADC2 constraints, not lab conditions.

*Demo Setup * 5-node K3s cluster representing autonomous platforms

Tactical mesh simulation:

2 Mbps bandwidth

20% packet loss

Variable latency

FleetLearn running without centralized cloud access

Demo Scenario

A new aerial threat appears mid-mission

Updated object detection model is committed to Git

Nodes pull updates opportunistically over the mesh

Canary deployment applies to 1–2 nodes

Network disruption and node loss are simulated

System automatically recovers and continues propagation

Full rollout completes after verification

*Measured Outcomes * 70%+ bandwidth reduction vs naive full-model transfer

<8 minutes to propagate updated intelligence

No fleet-wide disruption

Verified rollback on simulated failure

*Why This Didn’t Exist Before * GitOps platforms weren’t designed for mesh networks

Tactical AI pipelines focused on single device updates

Defense solutions optimized for control, not autonomy

Mesh radios matured faster than AI distribution tooling

*FleetLearn exists now because: * Kubernetes is finally viable at the edge

Tactical mesh is operationally deployed

AI is mission-critical, not experimental

*Strategic Impact * FleetLearn enables JADC2 systems to:

Adapt faster than adversaries

Maintain coherence under fragmentation

Push intelligence without centralized fragility

Treat AI as a living capability, not a static payload

One-Sentence Summary

FleetLearn is GitOps for JADC2—allowing AI intelligence to propagate safely and autonomously across the force, even when the network is denied.