Search and rescue (SAR) operations in mountainous terrain present significant challenges due to complex environments, time-critical decisions, and the need for effective human–AI collaboration. Existing approaches typically employ either fully autonomous systems that lack adaptability to varying task requirements, or fixed human–AI authority allocations that fail to leverage the distinct strengths of humans and AI across different mission phases. This paper proposes Phase-Aware Hierarchical Reinforcement Learning (PAHRL), a novel framework that dynamically allocates decision-making authority between human operators and AI agents based on identified task phases. First, we formulate the mountain SAR problem as a three-phase task structure: Wide Search (WS), Target Confirmation (TC), and Rescue Coordination (RC), and examine the consistency of this decomposition through unsupervised clustering analysis, supported by bootstrap stability (ARI = 0.983 ± 0.083) and multiple clustering metrics. Second, we design an adaptive authority mechanism with four levels (L1: Human-Led to L4: Full-Auto) that automatically adjusts human involvement based on current phase characteristics and environmental uncertainty estimates. Third, we introduce a priority-based task execution module that ensures efficient resource allocation across multiple rescue objectives while respecting authority constraints. Extensive experiments demonstrate that PAHRL outperforms baseline methods, achieving a 20.9% higher success rate compared to standard PPO (59.0% vs. 48.8%) and 66.7% improvement over heuristic approaches. PAHRL maintains 96.9% precision even under 60% noise conditions with only 0.09 false rescues per episode. Ablation studies further reveal that phase awareness serves as a critical robustness mechanism; removing phase detection causes complete mission failure under noisy conditions. These results evaluate that phase-aware dynamic authority allocation significantly enhances both efficiency and robustness in human–AI collaborative SAR missions. While demonstrated in a proof-of-concept simulation with computational human models, validation with real operators and more complex environments remains essential before operational deployment.

Very High Frequency (VHF) radio communication systems face significant challenges in modern electromagnetic environments, including spectrum congestion, dynamic interference, and varying channel conditions. Existing adaptive approaches rely on static rule-based switching or single-cycle optimization, which cannot accumulate operational experience across decision cycles. This paper proposes a digital twin-enabled online learning framework (DT-MAB) for adaptive waveform selection in tactical VHF communication. The framework employs a contextual multi-armed bandit algorithm (Lin-UCB) that continuously learns the mapping from channel conditions to optimal configurations, with the digital twin serving as a virtual exploration sandbox that screens candidate configurations before physical deployment—preventing link disruptions during exploratory actions. An expanded configuration space of 63 candidates (7 waveforms × 3 MAC protocols × 3 power levels) is constructed, and a hierarchical performance evaluation model combining voice quality, bit error rate, communication delay, and transmission range is developed using the Analytic Hierarchy Process (AHP) as the reward function for online learning. Experimental results across 10 random seeds demonstrate that DT-MAB achieves the lowest mean cumulative regret, reducing regret by 29% relative to MAB without a digital twin and by 16.5% relative to PSO-based optimization on average. Ablation experiments confirm that removing virtual exploration increases performance drop events by 49% (from 250 ± 79 to 373 ± 6), demonstrating that the digital twin is a functionally indispensable component of the online learning architecture.