Vision-and-language navigation (VLN) is a key task in Embodied AI, requiring agents to navigate diverse and unseen environments while following natural language instructions. Traditional approaches rely heavily on historical observations as spatio-temporal contexts for decision making, leading to significant storage and computational overhead. In this paper, we introduce MapNav, a novel end-to-end VLN model that leverages Annotated Semantic Map (ASM) to replace historical frames. Specifically, our approach constructs a top-down semantic map at the start of each episode and update it at each timestep, allowing for precise object mapping and structured navigation information. Then, we enhance this map with explicit textual labels for key regions, transforming abstract semantics into clear navigation cues and generate our ASM. MapNav agent using the constructed ASM as input, and use the powerful end-to-end capabilities of VLM to empower VLN. Extensive experiments demonstrate that MapNav achieves state-of-the-art (SOTA) performance in both simulated and real-world environments, validating the effectiveness of our method. Moreover, we will release our ASM generation source code and dataset to ensure reproducibility, contributing valuable resources to the field. We believe that our proposed Map-Nav can be used as a new memory representation method in VLN, paving the way for future research in this field.

This paper explores the embodied intelligence capabilities of GPT-4o, emphasizing its performance in key areas such as embodied reasoning, embodied manipulation, and embodied navigation. We are the first to systematically evaluate GPT-4o on embodied intelligence tasks. As an advanced language model, GPT-4o demonstrates significant potential in complex tasks by integrating visual and linguistic understanding, with particularly noteworthy abilities in image generation. In the embodied reasoning section, the study evaluates GPT-4o's performance in perception, spatial reasoning, temporal reasoning, planning, and causal reasoning. The findings reveal how the model extracts information from images to make informed inferences. The embodied manipulation section analyzes its applications in object understanding, environmental perception, and task planning, showcasing the model's adaptability in dynamic environments. Additionally, the embodied navigation section focuses on GPT-4o's ability to process navigation instructions, perform map reasoning, infer trajectories, and predict actions, indicating its effectiveness in navigation tasks. In summary, GPT-4o shows remarkable advancements in the field of embodied intelligence. Future development plans aim to further enhance its capabilities in multimodal interactions and complex environmental adaptability, laying a foundation for achieving higher levels of artificial intelligence. Through ongoing research and optimization, GPT-4o is poised to play a greater role in the practical applications of embodied intelligence.