Résumé
We present a multi-criteria navigation framework for mobile robots operating in complex, nuclear-like facilities. Heterogeneous layers—obstacles, radiation dose rate, and wireless connectivity—are harmonized into co-registered JSON/TIFF maps, normalized to a common [0, 1] scale, and fused into a pixelwise risk map via a weighted combination that encodes operator intent. Planning integrates a risk-aware A* with an exact Held–Karp stage for waypoint ordering, yielding interpretable control over distance, dose, and mission duration. A controlled evaluation across two canonical topologies (hallway versus shed), two obstacle densities, and three weight profiles demonstrates topology-dependent trade-offs: distance-focused weights shorten paths in corridors but increase dose, whereas safety-oriented weights substantially reduce dose in open spaces with modest penalties in length and time. The results show that a simple, transparent weighting scheme can steer mission behavior without modifying the planner, bridging perception and decision through map fusion. We discuss limitations related to static environments and scale and outline paths toward stochastic updates, chance constraints, and real-robot validation.