ObsGraph: Hierarchical Observation Representation for Embodied Reasoning and Exploration

2026-06-22

AI Summary

In complex and unfamiliar environments, robots struggle to efficiently acquire task-relevant information due to the absence of a unified scene representation and exploration mechanism. This work proposes ObsGraph—a hierarchical, observation-centric scene graph that organizes visual evidence into a three-level structure of rooms, viewpoints, and objects, thereby unifying scene representation, retrieval, and exploration. The approach innovatively integrates hierarchical graph structures with adaptive exploration, dynamically activating multi-scale exploration strategies—such as room-level navigation, viewpoint refinement, and boundary probing—based on evidence gaps identified in retrieval results. Experiments demonstrate that this method significantly improves both task success rates and exploration efficiency across multiple embodied reasoning and exploration benchmarks, validating the effectiveness of structured representations for goal-directed information gathering.

Abstract

Embodied reasoning and exploration are increasingly considered crucial abilities for robots operating in complex and unfamiliar environments. To accomplish tasks in such settings, an agent must identify and acquire the information necessary for the task through exploration. We propose ObsGraph, an observation-centric hierarchical scene graph that unifies scene representation, retrieval, and exploration. It retains visual evidence and organizes it into room-view-object layers: rooms provide coarse semantic anchors, views preserve contextual object covisibility, and objects store fine-grained details. On top of this representation, we perform coarse-to-fine hierarchical retrieval under a bounded budget, and crucially use retrieval outcomes to structure the exploration candidate space--activating room-level exploration, view refinement, or frontier exploration--thereby tightly coupling representation, retrieval, and adaptive multi-scale exploration. Experiments across embodied reasoning and exploration benchmarks demonstrate improved success and efficiency, highlighting the benefits of structured scene representation and more targeted information gathering driven by identified evidence gaps.

Problem

• embodied reasoning
• exploration
• scene representation
• hierarchical observation
• information gathering

Innovation

• hierarchical scene graph
• embodied reasoning
• adaptive exploration
• coarse-to-fine retrieval
• observation-centric representation