EVENT DOSSIER
A study published in September 2026 proposed Compact Bellman-Grounded Cognitive Maps, aimed at addressing the issue of cost perception in autonomous navigation. This study combined the Bellman equation from classical dynamic programming with cognitive maps to create a model capable of efficiently processing environmental states and evaluating action costs. By compressing the cognitive structure, this method enhances the ability to plan paths and optimize resources in complex environments, providing a new theoretical framework and technical solutions for low-cost, highly reliable autonomous navigation systems.
本文提出一种紧凑的贝尔曼接地认知地图（BCM），通过自监督贝尔曼接地目标和紧凑坐标编码，将可重用的认知地图建立在局部边成本之上，支持无需每目标重新训练的目标查询。在最大节点数为 1600 的加权网格上，BCM 保持完全成功率且与精确 Dijkstra 搜索的平均 Gap 仅为 5%，而基于连通性的谱基线约为 45%。随着图规模从 400 增长至 3600，其内存占用呈次线性增长并维持竞争性性能，展现出对复杂环境的可扩展性。该方法将加法路线成本写入紧凑、可重用的认知地图表示中，弥合了生物灵活性与最优路径规划之间的差距。