Feature Request: Investigate hierarchical AI architecture to improve RTS AI scalability and decision making

Hello,

I would like to propose investigating a hierarchical AI architecture for OpenRA's AI system.

The motivation comes from several common challenges in RTS AI architectures:

• As the number of units increases, updating and evaluating every agent independently can become increasingly expensive.
• Local unit-level decisions may not always maintain awareness of larger strategic goals.
• As AI behavior becomes more complex, adding more behaviors and rules can make the system harder to maintain and extend.
• Agents acting independently may repeat similar calculations instead of sharing higher-level context.

A possible approach to address these challenges is a hierarchical decision architecture.

I am developing an experimental architecture called HDDE (Hierarchical Deterministic Decision Engine):

https://github.com/Amlaach/HDDE-A-Zero-Allocation-Hierarchical-Utility-Architecture

The core idea is to separate AI responsibilities into layers:

• Commander layer:
  Responsible for strategic goals, planning, coordination and global decisions.

• Soldier layer:
  Responsible for local execution and immediate behaviors.

Additional concepts include:

• Time-sliced decision processing.
• Deterministic execution.
• Zero-allocation runtime design for predictable performance.

The goal is not to replace OpenRA's existing AI system, but to explore whether a hierarchical architecture could complement the current design and help with scalability, maintainability and strategic decision making.

Possible integration paths could include:

• A prototype AI module.
• An optional alternative AI architecture.
• Gradual integration of specific components where useful.

I would appreciate feedback from OpenRA developers and contributors:

• Does this approach address real challenges in the current AI architecture?
• Would such an architecture be compatible with OpenRA's existing design?
• Are there areas where this approach could be useful?

If the idea is considered relevant, I would be interested in contributing a proof-of-concept implementation or working on a PR with guidance from the community.

Thank you.

The issue names no OpenRA files, tests, or entry points. Start by reviewing OpenRA's existing AI system alongside the linked HDDE architecture, then determine whether the project needs a prototype, an optional alternative, or selected components. Done would require developer feedback and an agreed proof-of-concept scope.