Get Started
Home
Topics
Search
Library
Research questionHow can a navigation map encode heterogeneous route costs compactly while answering new goals without retraining?A reusable map must preserve nonuniform additive edge costs without storage that grows too quickly as the environment expands. It must also support new goal queries from one learned representation rather than rebuilding or retraining for each goal.
AI
AI Memory
Machine Learning
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.Compact Bellman-Grounded Cognitive Maps for Cost-Aware NavigationThe work studies BCM, a Bellman-grounded compact cognitive-map representation for weighted-grid navigation. Evidence covers success and path-cost gaps on graphs up to 1,600 nodes, plus memory and performance scaling from 400 to 3,600 nodes, with comparisons to exact Dijkstra search and a connectivity-based spectral baseline.research paper · Sep 4, 2026
Related questions
How can learned vehicle-routing solvers handle hard constraints without handcrafted heuristics or expensive exact optimization?How can closed-loop vision-language navigation learn effectively despite distribution shift and sparse micro-action rewards?How can embodied vision-language navigation follow instructions in unseen environments with limited data and memory?How can we find minimal road-network changes that make a user-selected route optimal?