Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 02, 2018 - June 05, 2018
This research aims to propose a method to build a semantic map to be used for a robot path planning in a greenhouse. Existing mapping methods only consider whether there are obstacles in a certain region. They are not sufficient for path planning in greenhouses where paths are often covered by branches and leaves which are also recognized as obstacles. We propose a mapping method which generates a map with semantic information on the types of obstacles. By integrating 3D mapping function provided by RTAB-Map, an RGB-D based visual SLAM (Simultaneous Localization And Mapping) method, and semantic segmentation by SegNet, a deep convolutional encoder-decoder architecture for semantic pixel-wise labeling, we obtain a 3D map with semantic labels. In order to deal with uncertainty of observations, we introduce a probabilistic label update strategy. We voxelize the map and calculate the probability of each label of each voxel by voting. In addition, using the fact that the robot traverses a voxel, the probabilities in the voxel are updated using Bayes’ rule. Through evaluations, we confirmed that the proposed method can perform a more accurate semantic labeling than the one only using SegNet.
