Kaiqi Huang

Existing LLM-based agents primarily utilize coarse-grained experiential memory, where experiences are retrieved based on global task or scene context. While effective in simple settings, such coarse-grained memory lacks the situational alignment required for complex multi-step decision-making. As...