Construction of a Lightweight Simulation Environment Based on Topological Mapping

Autonomous mobile robots require safe and efficient environmental representations for traversability-aware navigation and learning. In particular, self-learning of traversability from real-world driving experience can require robots to operate near the limits of their mobility, which introduces physical risks such as falls, malfunctions, or hardware damage. To provide a safer complementary environment for such validation and future learning, this paper proposes ATC-Mesh. This framework constructs a lightweight, simulation-ready mesh environment from the topological map produced by Adaptive Resonance Theory-based Topological Clustering with Different Topologies (ATC-DT) using high-density LiDAR point clouds. Unlike standard mesh reconstruction methods that directly process dense point clouds, ATC-Mesh generates a compact mesh from the node–edge structure of the topological map while preserving traversability attributes associated with the topological map. Experiments using two real-world LiDAR datasets show that ATC-Mesh achieves competitive mesh-construction quality compared with standard baseline methods while reducing construction time and mesh size. Gazebo experiments further show that the generated meshes can support waypoint-based navigation with a low simulation load.