Endowing Embodied Agents with Spatial Reasoning Capabilities for Vision-and-Language Navigation

Researchers introduce BrainNav, a bio-inspired navigation framework that mimics biological spatial cognition to enhance Vision-and-Language Navigation in mobile robots. The system addresses spatial hallucination issues when transferring from simulation to real-world environments, demonstrating superior performance in zero-shot real-world testing.

• →BrainNav framework uses dual-map and dual-orientation strategies inspired by biological spatial cognition theories.
• →The system includes five core modules that mimic brain functions including hippocampal memory and visual cortex perception.
• →Framework successfully reduces spatial hallucinations that occur when transferring simulated capabilities to real-world scenarios.
• →BrainNav outperforms existing SOTA VLN-CE methods in zero-shot real-world testing without requiring fine-tuning.
• →The system demonstrates compatibility with GPT-4 and validates effectiveness using Limo Pro robot in laboratory environments.