Topological Mapping and Navigation using a Monocular Camera based on AnyLoc

TL;DR

This work presents a monocular-camera approach for topological mapping and goal-directed navigation by leveraging AnyLoc to construct a directed graph of key nodes and relations without relying on metric maps. The Mapping component builds and optimizes a sparse topological map through image-similarity thresholds, loop closures, and node management, while the Navigation component localizes, plans via Dijkstra, and controls motion using image-based decisions. The method demonstrates robust performance across real and simulated environments without pre-training, achieving a substantial improvement over ResNet-based baselines and revealing critical insights about map sparsity. The practical impact lies in lightweight, fast-deployable visual navigation suitable for robots and humans in dynamic, unknown environments.

Abstract

This paper proposes a method for topological mapping and navigation using a monocular camera. Based on AnyLoc, keyframes are converted into descriptors to construct topological relationships, enabling loop detection and map building. Unlike metric maps, topological maps simplify path planning and navigation by representing environments with key nodes instead of precise coordinates. Actions for visual navigation are determined by comparing segmented images with the image associated with target nodes. The system relies solely on a monocular camera, ensuring fast map building and navigation using key nodes. Experiments show effective loop detection and navigation in real and simulation environments without pre-training. Compared to a ResNet-based method, this approach improves success rates by 60.2% on average while reducing time and space costs, offering a lightweight solution for robot and human navigation in various scenarios.

Figures (11)

• Figure 1: Pipeline of the proposed method. From topological map building to goal navigation. There are two main components: (1) Mapping and (2) Navigation.
• Figure 2: Node information. The left side shows a simplified topological map, while the right side displays the information contained within each node.
• Figure 3: Topological mapping process. Starting from $N_0$ and moving towards $N_{90}$ in a clockwise direction, the process of node addition, distance calculation, and loop closure detection is described.
• Figure 4: Topological map node optimization. In case (a), $N_1$ is close to both $N_0$ and $N_2$. In case (b), nodes $N_1$ and $N_4$ are similar. The right side of the image shows the optimized results.
• Figure 5: Global navigation system overview. (a) Locate nodes $N_{75}$ and $N_0$ in the graph, corresponding to the current observation and the target image, respectively. (b) The shortest path on the graph between these nodes is selected. (c) Node selection is performed during the movement towards the goal.