Centralized Training, Decentralized Execution/ CTDE

Definition

Centralized Training, Decentralized Execution is training pattern where agents learn with shared global information but deploy using local observations. In defense applications, it produces policies that coordinate well during training yet survive disconnected field execution. The hard part is training-deployment mismatch and hidden dependence on unavailable global state, especially when systems are deployed across contested links, coalition boundaries, and mixed human-machine teams. KhanBMS treats it as a fit for KhanBMS formations that rehearse as a Tumen but fight as local Arbans, tying the concept back to modular command, edge execution, and auditable authority.

Reference attributes

Layer

MARL training pattern

Operational value

Produces policies that coordinate well during training yet survive disconnected field execution

Primary risk

Training-deployment mismatch and hidden dependence on unavailable global state

KhanBMS role

A fit for KhanBMS formations that rehearse as a Tumen but fight as local Arbans

Related terms

• Multi-Agent Reinforcement Learning (MARL)
  Reinforcement-learning framework where multiple agents learn cooperative or adversarial behavior together.
• Cooperative Perception
  Shared perception where multiple platforms combine local observations to improve detection and tracking.
• Federated Learning (FL)
  Training approach where nodes learn from local data and share updates instead of raw data.