https://github.com/7yloo/Q-MAS-Quantum-inspired-Multi-Agent-Swarm

language: - en license: mit library_name: sklearn tags: - swarm-intelligence - quantum-inspired - multi-agent-systems - reinforcement-learning - distance-prediction - iraq - independent-research pipeline_tag: reinforcement-learning

🐜 Q-MAS Distance Predictor

Model Description

This is the trained MLP predictor for Layer 7 of the Q-MAS framework. It estimates the Euclidean distance from an agent's current position to the target based on 10 environmental and historical features.

Attribute	Value
Architecture	10-32-16-8-1
Training samples	50
MAE	0.055 (normalized distance)
Input features	10
Output	Normalized distance (0-1)

Performance

When integrated with the 6-layer Q-MAS baseline:

Metric	6-Layer	7-Layer	Improvement
Survivors/10	2.6 ± 0.8	3.6 ± 0.9	+38%
First target (steps)	55.0 ± 8.2	52.2 ± 9.1	-2.8 steps
Hazard violations	0	0	0%

✅ 5 independent runs · Zero hazard violations · p < 0.05

How to Use

import joblib
import numpy as np

# Load model and scaler
model = joblib.load("qmas_predictor.pkl")
scaler = joblib.load("qmas_scaler.pkl")

# 10 features: [x/10, y/10, signal, pheromone_mean, pheromone_max,
#               visit_mean, visit_max, time/100, epsilon, history/20]
features = np.array([0.5, 0.5, 2.5, 0.3, 0.8, 1.2, 3.0, 0.4, 0.5, 0.3]).reshape(1, -1)

# Predict distance
features_scaled = scaler.transform(features)
pred_distance = model.predict(features_scaled)[0]
actual_distance = pred_distance * 15.0  # Convert to grid units (0-15)

print(f"Predicted distance to target: {actual_distance:.1f} units")