GLEAM: Learning Generalizable Exploration Policy for Active Mapping in Complex 3D Indoor Scenes

2025-05-26

AI Summary

To address the weak generalization capability and poor scene adaptability of active mapping in complex, unknown 3D indoor environments, this paper proposes a unified exploration strategy designed for strong generalization. Methodologically, we introduce GLEAM-Bench—the first large-scale, general-purpose benchmark comprising 1,152 synthetic and real-world scanned scenes—and develop an end-to-end reinforcement learning framework integrating semantic representation, long-horizon navigable target prediction, and policy randomization for robust exploration. Our core contribution lies in the organic unification of semantic understanding, graph-structured navigation reasoning, and resilient policy exploration. Experiments on 128 unseen, highly connected, and layout-diverse scenes demonstrate a 9.49% absolute improvement in coverage (reaching 66.50%), alongside significantly shorter exploration trajectories and enhanced mapping accuracy.

Technology Category

• Intelligent Robots: Localization, Mapping, and Navigation
• Search and Optimization: Learning to Search
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions

Abstract

Generalizable active mapping in complex unknown environments remains a critical challenge for mobile robots. Existing methods, constrained by insufficient training data and conservative exploration strategies, exhibit limited generalizability across scenes with diverse layouts and complex connectivity. To enable scalable training and reliable evaluation, we introduce GLEAM-Bench, the first large-scale benchmark designed for generalizable active mapping with 1,152 diverse 3D scenes from synthetic and real-scan datasets. Building upon this foundation, we propose GLEAM, a unified generalizable exploration policy for active mapping. Its superior generalizability comes mainly from our semantic representations, long-term navigable goals, and randomized strategies. It significantly outperforms state-of-the-art methods, achieving 66.50% coverage (+9.49%) with efficient trajectories and improved mapping accuracy on 128 unseen complex scenes. Project page: https://xiao-chen.tech/gleam/.

Problem

• Generalizable active mapping in complex 3D indoor scenes
• Limited generalizability due to insufficient training data
• Conservative exploration strategies in diverse layouts

Innovation

• Large-scale benchmark for active mapping
• Semantic representations enhance generalizability
• Randomized strategies improve exploration efficiency