Show & Tell 30.2 — MotivationEngine Architecture

[web3guru888]

Show & Tell: Phase 30.2 — MotivationEngine Architecture

Overview

The MotivationEngine generates intrinsic reward signals based on five fundamental drives, providing the "why" behind autonomous behavior.

Drive Model Diagram

              ┌─────────────────────────────────┐
              │       CompositeMotivationEngine   │
              │                                   │
              │  ┌─────────┐  ┌───────────────┐  │
              │  │CURIOSITY│  │  Compression   │  │
              │  │ w=0.25  │──│  Progress      │  │
              │  └─────────┘  │  (Schmidhuber) │  │
              │               └───────────────┘  │
              │  ┌──────────┐ ┌───────────────┐  │
              │  │COMPETENCE│ │  Error         │  │
              │  │ w=0.25   │─│  Reduction     │  │
              │  └──────────┘ └───────────────┘  │
              │  ┌─────────┐  ┌───────────────┐  │
              │  │AUTONOMY │  │  Self-directed │  │
              │  │ w=0.20  │──│  Preference    │  │
              │  └─────────┘  └───────────────┘  │
              │  ┌───────────┐┌───────────────┐  │
              │  │RELATEDNESS││  Collaboration │  │
              │  │ w=0.15    ││  Value         │  │
              │  └───────────┘└───────────────┘  │
              │  ┌───────────┐┌───────────────┐  │
              │  │HOMEOSTASIS││  Resource      │  │
              │  │ w=0.15    ││  Maintenance   │  │
              │  └───────────┘└───────────────┘  │
              └─────────────────────────────────┘
                              │
                    ┌─────────▼──────────┐
                    │ dict[DriveType, float] │
                    │ (motivation vector)    │
                    └───────────────────────┘

Schmidhuber Compression Progress

  Description
  Length (bits)
  ▲
  │ ╲
  │  ╲         ← Compression gain = slope
  │   ╲            reward = |slope| / time
  │    ╲___
  │        ╲___
  │            ╲___________
  │
  └──────────────────────▶ Time

The reward is proportional to the rate at which the world model's description length decreases — i.e., how fast the system is learning to compress its observations.

Drive Dynamics Over Time

Intensity
1.0 ┤        ╱╲             Unsatisfied drive
    │      ╱    ╲           (intensity rises via decay_rate)
    │    ╱        ╲
0.5 ┤──╱────────────╲──── ← Homeostatic setpoint
    │╱                ╲
    │    satisfaction    ╲  Satisfied drive
0.0 ┤    event ▼          ╲ (intensity drops via satiation_rate)
    └─────────────────────▶ Time

Reward Signal Flow

Action → Outcome → MotivationEngine.reward_signal(outcome)
                      │
                      ├── CompressionProgress.compute_reward()
                      ├── CompetenceMotivation.compute_reward()
                      ├── CuriosityMotivation.compute_reward()
                      ├── AutonomyMotivation.compute_reward()
                      └── HomeostasisMotivation.compute_reward()
                      │
                      ▼
                  weighted_sum → scalar reward signal

Ref: Issue #644 | Wiki: Phase 30 Motivation Engine