Planning Loops

Planning is the process of generating a sequence of actions before executing them. In the real world, plans often fail, so the AGI must continuously monitor and adjust.

The Planning Cycle

A planning loop in PRIMUS involves three main stages: Projection, Execution, and Monitoring.

graph LR
    Plan[Plan Formulation] -->|Project| Simulate[Mental Simulation]
    Simulate -->|Is Viable?| Execute[Action Execution]
    Execute -->|Observe| Monitor[Performance Monitoring]
    Monitor -->|Mismatch?| Plan

1. Projection (Simulation)

The agent uses its "World Model" to imagine what would happen if it took certain actions. It doesn't act; it stays in its "head."

2. Execution

Once a viable plan is found, the agent sends commands to its actuators.

3. Monitoring (The Feedback Loop)

This is the most critical part. As the agent acts, it compares what actually happens with what it projected would happen.

• Success: Continue the plan.
• Minor Error: Adjust the plan (e.g., steer slightly left).
• Major Error: Stop and re-plan from scratch.

Hierarchical Planning

AGI doesn't plan every muscle movement in its head. It plans at multiple levels of abstraction.

• High-Level: "Go to the supermarket."
• Mid-Level: "Walk to the front door."
• Low-Level: "Lift left leg."

This hierarchical approach allows the agent to be efficient—it doesn't need to re-think its entire shopping trip just because it tripped over a rug.

Next: Decision Making