GUIDE: A Diffusion-Based Autonomous Robot Exploration Framework Using Global Graph Inference

TL;DR

This work proposes GUIDE, a novel exploration framework that synergistically combines global graph inference with diffusion-based decision-making, and introduces a region-evaluation global graph representation that integrates both observed environmental data and predictions of unexplored areas.

Abstract

Autonomous exploration in structured and complex indoor environments remains a challenging task, as existing methods often struggle to appropriately model unobserved space and plan globally efficient paths. To address these limitations, we propose GUIDE, a novel exploration framework that synergistically combines global graph inference with diffusion-based decision-making. We introduce a region-evaluation global graph representation that integrates both observed environmental data and predictions of unexplored areas, enhanced by a region-level evaluation mechanism to prioritize reliable structural inferences while discounting uncertain predictions. Building upon this enriched representation, a diffusion policy network generates stable, foresighted action sequences with significantly reduced denoising steps. Extensive simulations and real-world deployments demonstrate that GUIDE consistently outperforms state-of-the-art methods, achieving up to 18.3% faster coverage completion and a 34.9% reduction in redundant movements.

Figures (8)

• Figure 1: Exploration in a structured indoor environment with a mobile robot using GUIDE. Left: the robot’s actual position in the real world. Top right: the robot leverages the global graph representation to generate exploration trajectories, where blue nodes indicate predictions of unexplored areas and the red line denotes the generated action sequence. Bottom right: the 3D occupancy map constructed by the robot during exploration.
• Figure 2: Overview of the proposed GUIDE framework with three modules: (Sec. IV-A) Environmental Extraction updates the observation, samples free nodes, and decomposes the whole space; (Sec. IV-B) Region-Evaluation Graph Inference leverages observations to predict nodes and filters them for reliability; (Sec. IV-C) Diffusion-Based Decision conditions a diffusion policy on the global graph to generate stable long-horizon trajectories.
• Figure 3: Illustration of the process of node prediction with region-evaluation. (a) Observed nodes (free and obstacles) as input to the predictor. (b) Predicted free nodes $\mathcal{V}_p$ from the Global Node Predictor. (c) Final unknown-node set $\mathcal{V}_u$ after region-evaluation, where high-score regions retain predictions while others are replaced by centroids or discarded.
• Figure 4: Illustration of (a) utility update, where nodes with the same frontier count obtain different utilities depending on predicted nodes within the utility update range; and (b) edge construction, showing connections between different node types.
• Figure 5: Action sequence on the same map at approximately the same locations (a)–(c). In each pair, the bottom shows diffusion based only on observed information and the top our method, with red indicating the generated action sequence and green the region-evaluated unknown nodes. By leveraging the global graph representation, our method produces more reasonable trajectories.