Semantic Environment Atlas for Object-Goal Navigation

Nuri Kim,Jeongho Park,Mineui Hong,Songhwai Oh

DOI: https://doi.org/10.1016/j.knosys.2024.112446

2024-10-05

Abstract:In this paper, we introduce the Semantic Environment Atlas (SEA), a novel mapping approach designed to enhance visual navigation capabilities of embodied agents. The SEA utilizes semantic graph maps that intricately delineate the relationships between places and objects, thereby enriching the navigational context. These maps are constructed from image observations and capture visual landmarks as sparsely encoded nodes within the environment. The SEA integrates multiple semantic maps from various environments, retaining a memory of place-object relationships, which proves invaluable for tasks such as visual localization and navigation. We developed navigation frameworks that effectively leverage the SEA, and we evaluated these frameworks through visual localization and object-goal navigation tasks. Our SEA-based localization framework significantly outperforms existing methods, accurately identifying locations from single query images. Experimental results in Habitat scenarios show that our method not only achieves a success rate of 39.0%, an improvement of 12.4% over the current state-of-the-art, but also maintains robustness under noisy odometry and actuation conditions, all while keeping computational costs low.

Artificial Intelligence,Robotics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to improve the target - object navigation ability of agents (such as robots) in unknown environments. Specifically, the paper introduces a new mapping method - Semantic Environment Atlas (SEA), aiming to enhance vision - based navigation capabilities. SEA enriches the context information of navigation by constructing and using semantic maps to depict in detail the relationships between locations and objects. This method can not only handle the challenges brought by sensor noise, but also predict the positions of objects in partially observed maps, and can adapt to environmental changes and self - update to adjust target positions or explore alternative target locations. In addition, SEA can improve the robustness of navigation through semantic path planning without using global pose sensors while maintaining low computational costs.

The paper verifies the effectiveness of SEA through experiments, especially in the object - goal navigation tasks in the Habitat simulator. The SEA method not only significantly improves the success rate to 39.0%, which is 12.4% higher than the existing state - of - the - art methods, but also maintains robust performance under noisy odometry and execution conditions. This indicates that SEA has high practical value and potential in practical applications.