RouteFormer: Transformer-RL Framework for Autonomous Vehicle Routing

Researchers propose RouteFormer, a novel framework combining transformer self-attention with reinforcement learning for single-agent routing in graph-based terrains. Designed for autonomous surveillance in IoT networks, it addresses NP-hard combinatorial optimization problems without labeled training data. Evaluated on varying graph sizes simulating reconnaissance missions, the model handles complex task dependencies and resource availability. The framework adapts to dynamic environments, overcoming limitations of conventional heuristics. Results show effectiveness for missions requiring multiple action profiles.