Mapping of unknown indoor environments for 3D reconstruction and developing virtual reality tours are popular in real-estate industries. However, the task is tedious for the operator, and error-prone without real-time visualization. Therefore, using a robot to autonomously explore and map the environment is beneficial.
Considering the mapping of Dutch houses, the task involves identifying locations to capture 360-degree images of the rooms and detecting regions that need to be visited to map the complete house in minimum time. However, traditional exploration approaches using 2D LiDAR are not explicitly designed for detecting locations suitable for scanning.
The designed exploration algorithm locates points in each room in which the scanning has to be performed and plan a path to explore more such locations. The proposed algorithm uses inspiration from human’s spatial cognition to develop a non-linear regression model of the environment using Gaussian processes. This active learning-based framework is used to identify locations suitable for scanning.
