Towards Dual-Brain Minimal Sufficient Representation for Vision-Language Navigation

2026-07-25

AI Summary

Vision-and-language navigation often suffers from poor generalization and high computational cost due to the retention of excessive task-irrelevant details. This work proposes BrainNav, a novel framework that introduces minimal sufficient representations to this domain for the first time. BrainNav employs a dual-brain architecture comprising a logic-anchored model, a minimalist constraint alignment module, and a low-rank compressed world model to jointly align discrete linguistic intentions with continuous latent dynamics. This approach enables efficient cross-modal alignment while preserving robustness and computational efficiency. Evaluated on the R2R-CE and RxR-CE benchmarks, BrainNav achieves significant improvements of 2.0%/1.0% and 0.94%/0.78% in Success Rate (SR) and Success weighted by Path Length (SPL), respectively, on the val-unseen splits, outperforming current state-of-the-art methods.

Abstract

Vision-and-Language Navigation in continuous environments (VLN-CE) requires an agent to ground language in egocentric observations and plan in unseen scenes. Although recent multimodal large models and world-model-based methods have improved navigation, they often preserve excessive task-irrelevant detail, weakening generalization and increasing computational burden. We propose BrainNav, a navigation framework grounded in the Principle of Minimal Sufficiency. BrainNav consists of three components: a Logical Anchor Model that implements instruction-aware selective perception to suppress environmental noise, a Minimalist Constraint Alignment module that serves as a compact cross-modal bottleneck, efficiently synchronizing discrete linguistic intent with continuous latent dynamics while filtering out redundant information, and a Compression World Model that predicts action-conditioned states within a condensed, low-rank latent space. These modules align semantic intent with spatial perception, enhancing the agent's robustness and efficiency in complex tasks. Experiments show that BrainNav improves over prior SOTA by 2.0 % / 1.0 in SR/SPL on R2R-CE val-unseen and 0.94 % / 0.78 on RxR-CE val-unseen. These results indicate that minimally sufficient world representations provide an effective foundation for robust VLN.

Problem

• Vision-Language Navigation
• Minimal Sufficiency
• Task-Irrelevant Detail
• Generalization
• Computational Burden

Innovation

• Minimal Sufficient Representation
• Vision-Language Navigation
• Cross-modal Bottleneck
• World Model Compression
• Selective Perception