一句话：提出决策感知信息共享 DAIS，以决策改进价值信息奖励探索性感知，提升多智能体强化学习游戏测试覆盖率。
现代3D游戏规模、开放性与交互复杂度不断提升，自动测试愈发重要。多智能体强化学习可并行探索，但协作场景面临部分可观测、非平稳性和信用分配等挑战。论文在Lab Recruits的主动-被动协作测试基础上提出决策感知信息共享（DAIS），一种基于决策改进的信息价值（VoI）架构：将探索性感知的奖励与主动智能体决策的可度量改进挂钩。初步实验表明，DAIS相对标准主动-被动设置提升了覆盖率，而后者本身优于单智能体基线。
Automated testing of 3D games is increasingly important as modern titles grow in scale, openness, and interaction complexity. Multi-agent reinforcement learning (MARL) offers parallelized exploration, but collaborative settings introduce partial observability, non-stationarity, and credit assignment challenges. Building on active-passive collaborative testing in Lab Recruits, we propose Decision-Aware Information Sharing (DAIS), a Decision-Improvement Value-of-Information (VoI) architecture that rewards exploratory sensing based on its measurable improvement to the active agent's decision-making. Preliminary experiments show that DAIS improves coverage relative to the standard active-passive setup, which itself outperforms a single-agent baseline.