Abstract:Exploration is a critical function for autonomous mobile robots. Traditionally, the entire map has to be processed to extract frontiers and perform path planning. However, as the robot explores the environment, the map grows over time, and increasing computational resources are required, especially for large-scale environments. Moreover, only a few methods focus on the exploration on point cloud maps. Here, I present a new practical method to autonomous mobile robot exploration based on a sparse, relatively small-size point cloud local map, which combines Rapidly-exploring Random Tree (RRT) and dynamic window approach (DWA) algorithm together. The local map is built from the consecutive inputs of raw point clouds using an inexpensive 3D sensor, i.e. Kinect V2. Frontiers are effectively detected and local path planning is performed by RRT algorithm directly on unordered point cloud local maps. Motion planning is performed online by DWA to avoid obstacles and direct a nonholonomic mobile robot towards frontiers separating known environments from unknown environments. Embedded with simultaneous localization and mapping (SLAM) system of my previous research, the performance of the proposed method is evaluated in a large-scale customized virtual environment with a size of $33 \times 29 \times 6\mathrm{m}$ using Gazebo as the robotic simulator. The results suggest that the proposed algorithm can accurately direct the nonholonomic mobile robot to unexplored environments in real time. Also, it successfully helps build a coherent semi-metric topological map. The proposed algorithm shows great efficient performance and is suitable for both static and dynamic environments. Combination of RRT and DWA algorithm on point clouds, as a general approach, can be extended to generic 3D nonplanar physical environments. Videos of the experiments can be found at https://youtu.be/0i766fhs9Ds.

GPU Accelerated Real-Time Traversability Mapping.

GEM: Online Globally Consistent Dense Elevation Mapping for Unstructured Terrain

Real-time Localization, Mapping and Navigation for Quadruped Robots

VDBblox: Accurate and Efficient Distance Fields for Path Planning and Mesh Reconstruction

Mapbuilding for Dynamic Environments Using Grid Vectors

Adaptive Block Elevation Mapping for Large-scale Scene.

Mobile Robot Exploration Based on Rapidly-exploring Random Trees and Dynamic Window Approach

RenderMap: Exploiting the Link Between Perception and Rendering for Dense Mapping

Real-Time Elevation Mapping with Bayesian Ground Filling and Traversability Analysis for UGV Navigation

An Approach for Multi-Objective Obstacle Avoidance Using Dynamic Occupancy Grid Map

Dense mapping from an accurate tracking SLAM

Gaussian Process-based Traversability Analysis for Terrain Mapless Navigation

Rapid Traversability Assessment In 2.5d Grid-Based Map On Rough Terrain

3D Traversability Map Generation for Mobile Robots Based on Point Cloud

Traversable Map Construction and Robust Localization for Unstructured Road Environments1

Real-Time Metric-Semantic Mapping for Autonomous Navigation in Outdoor Environments

A Lightweight, Centralized, Collaborative, Truncated Signed Distance Function-Based Dense Simultaneous Localization and Mapping System for Multiple Mobile Vehicles

Semantic Probabilistic Traversable Map Generation For Robot Path Planning

An improved Gmapping algorithm based map construction method for indoor mobile robot

Fully Automatic Large-Scale Point Cloud Mapping for Low-Speed Self-Driving Vehicles in Unstructured Environments

Highly Dynamic Visual SLAM Dense Map Construction Based on Indoor Environments