2D LiDARデータの補完機能付きVAEを用いた移動ロボットの自己位置同定手法

抄録

We propose AT-COMPL-VAE (VAE with autocompletion feature), which is an extension of VAE (variational autoencoder) to complement 2D LiDAR data and extract features suitable for self-localization. AT-COMPL-VAE is a multi-task neural network with three neural networks: autocompletion, which completes the missing 2D LiDAR data; encoder, which compresses the input data; and decoder, which restores the original input data from the compressed data. In the simulation experiments conducted to verify the effectiveness of the proposed method, we evaluated the performance of 2D LiDAR data complementation and the performance of self-localization using the complemented data. It is demonstrated that the data complemented by the proposed method is close to the true value compared to the data complemented by random values and that the success rate of self-localization using the complemented data was improved by more than 80% compared to the case using the data with missing data due to occlusion.