Study of the Path Planning Based on SLAM and Semantic Segmentation by Deep Learning for Forest Environmental Monitoring Robot.

Abstract

In recent years, we are aware of many environmental problems, such as environmental destruction, reduction of wild animals. It is required to make use of autonomous robots for environmental monitoring to predict disasters and grasp situations. In this research, we consider image processing technology that contributes to environmental monitoring by WAMOT (Waseda Animal Monitoring Robot), an environmental monitoring robot developed for creating environment and animal ecology maps. WAMOT uses a charging station to withstand long-term autonomous observation. When the battery of the robot runs short, it automatically returns to the charging station. As soon as charging is completed, repeat the operation of starting the observation again to grasp the surrounding environment. In this research, we aim to make the Path Planning Based on SLAM and Semantic Segmentation by Deep Learning. We propose a method the path planning that based on SLAM and cost map reflecting the result of Semantic Segmentation. SLAM in irregular areas like forests is estimating the robot's state from the stable self position estimation and combining wheel odometry. Cost map based on the environment map (self position estimation result) and the estimation result by the deep learning from RGB image. As a result of the experiment, path planning from cost map generation reflecting the result of environment estimation could be performed.