PRISM-TopoMap: Online Topological Mapping with Place Recognition and Scan Matching

TL;DR

PRISM-TopoMap introduces an online topological mapping framework that builds and maintains a graph of locally aligned locations without relying on global metric coordinates. It combines a multimodal place recognition model (MSSPlace-G) with a 2D feature-based scan matching refinement and a rule-based graph maintenance module to ensure connectivity and reduce false links. The approach demonstrates superior computational efficiency and competitive mapping quality across photorealistic simulations and a real-robot experiment, with public code available. This work offers a scalable and drift-resilient alternative to dense SLAM for long-range indoor navigation and planning.

Abstract

Mapping is one of the crucial tasks enabling autonomous navigation of a mobile robot. Conventional mapping methods output a dense geometric map representation, e.g. an occupancy grid, which is not trivial to keep consistent for prolonged runs covering large environments. Meanwhile, capturing the topological structure of the workspace enables fast path planning, is typically less prone to odometry error accumulation, and does not consume much memory. Following this idea, this paper introduces PRISM-TopoMap -- a topological mapping method that maintains a graph of locally aligned locations not relying on global metric coordinates. The proposed method involves original learnable multimodal place recognition paired with the scan matching pipeline for localization and loop closure in the graph of locations. The latter is updated online, and the robot is localized in a proper node at each time step. We conduct a broad experimental evaluation of the suggested approach in a range of photo-realistic environments and on a real robot, and compare it to state of the art. The results of the empirical evaluation confirm that PRISM-Topomap consistently outperforms competitors computationally-wise, achieves high mapping quality and performs well on a real robot. The code of PRISM-Topomap is open-sourced and is available at: https://github.com/kirillMouraviev/prism-topomap.

Figures (7)

• Figure 1: A graph of locations generated from the sensory inputs. Different colors mark different locations, blue lines denote transitions between them.
• Figure 2: A scheme of the proposed PRISM-TopoMap method which takes multi-camera images and point cloud as input. It includes $F_{PR}$ place encoder, scan matching module, and graph maintaining module. The output is the graph of locations $G_t$ at the moment $t$.
• Figure 3: A scheme of the graph maintaining module: checking that the robot is inside $v_{cur}^{t-1}$, changing $v_{cur}$ according to edges and localization results, and addition of new location.
• Figure 4: A scheme of the modified MSSPlace place recognition method, referred to as MSSPlace-G.
• Figure 5: Results of scan matching for all the evaluated methods, compared with ground truth matching. The IoU between scans is 0.65.