MapDream: Task-Driven Map Learning for Vision-Language Navigation
This addresses the challenge of efficient navigation for agents in partially observed 3D environments by learning maps optimized for navigation rather than reconstruction, representing an incremental improvement over existing methods.
The paper tackles the problem of Vision-Language Navigation by proposing MapDream, a framework that learns compact, task-driven map representations through autoregressive bird's-eye-view image synthesis, achieving state-of-the-art monocular performance on R2R-CE and RxR-CE benchmarks.
Vision-Language Navigation (VLN) requires agents to follow natural language instructions in partially observed 3D environments, motivating map representations that aggregate spatial context beyond local perception. However, most existing approaches rely on hand-crafted maps constructed independently of the navigation policy. We argue that maps should instead be learned representations shaped directly by navigation objectives rather than exhaustive reconstructions. Based on this insight, we propose MapDream, a map-in-the-loop framework that formulates map construction as autoregressive bird's-eye-view (BEV) image synthesis. The framework jointly learns map generation and action prediction, distilling environmental context into a compact three-channel BEV map that preserves only navigation-critical affordances. Supervised pre-training bootstraps a reliable mapping-to-control interface, while the autoregressive design enables end-to-end joint optimization through reinforcement fine-tuning. Experiments on R2R-CE and RxR-CE achieve state-of-the-art monocular performance, validating task-driven generative map learning.