MUVLA: Learning to Explore Object Navigation via Map Understanding

2025-09-30

AI Summary

This paper addresses the inefficiency and poor generalization of exploration strategies for embodied agents navigating toward target objects in unknown environments. We propose a unified exploration framework grounded in semantic map understanding. Methodologically, we introduce a three-stage training paradigm: (1) constructing a structured semantic map that fuses visual, linguistic, and spatial contextual information; (2) incorporating a return-value prediction module to model dense short-horizon rewards, thereby strengthening supervision signals; and (3) jointly optimizing the action policy end-to-end via imitation learning and reinforcement learning. Our key contribution lies in explicitly encoding historical observations and spatial relationships into a learnable, reasoning-capable map representation, enabling robust learning from heterogeneous demonstrations—including partial successes and failures. Evaluated on HM3D and Gibson benchmarks, our approach achieves significant improvements in exploration efficiency and cross-scene generalization.

Technology Category

• Search and Optimization: Learning to Search
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Intelligent Robots: Embodied AI

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

In this paper, we present MUVLA, a Map Understanding Vision-Language-Action model tailored for object navigation. It leverages semantic map abstractions to unify and structure historical information, encoding spatial context in a compact and consistent form. MUVLA takes the current and history observations, as well as the semantic map, as inputs and predicts the action sequence based on the description of goal object. Furthermore, it amplifies supervision through reward-guided return modeling based on dense short-horizon progress signals, enabling the model to develop a detailed understanding of action value for reward maximization. MUVLA employs a three-stage training pipeline: learning map-level spatial understanding, imitating behaviors from mixed-quality demonstrations, and reward amplification. This strategy allows MUVLA to unify diverse demonstrations into a robust spatial representation and generate more rational exploration strategies. Experiments on HM3D and Gibson benchmarks demonstrate that MUVLA achieves great generalization and learns effective exploration behaviors even from low-quality or partially successful trajectories.

Problem

• Develops a vision-language-action model for object navigation tasks
• Uses semantic maps to structure spatial context and historical information
• Amplifies supervision through reward modeling for effective exploration strategies

Innovation

• Map Understanding Vision-Language-Action model for object navigation
• Leverages semantic map abstractions to structure historical information
• Employs three-stage training with reward amplification strategy