SLAM for Robotic Navigation by Fusing RGB-D and Inertial Data in Recurrent and Convolutional Neural Networks

摘要: Simultaneous localization and mapping (SLAM) is a key component for mobile robot navigation that enables many service robotic applications. The capacity of acquiring accurate 3D-map an environment critical robots to perform various tasks with high degree autonomy. Due the indoor complexity sensor uncertainties, SLAM remains challenging task in domain 3D reconstruction. In this paper, we propose simple yet effective solution RGB-D based by integrating Inertial Measurement Unit (IMU) into recurrent convolutional neural network leads enhanced pose estimation point cloud registration. IMU data provide advantage fast rate inertial measurement drift error reduction. Specifically, imposing additional constraints from device, optimal long-short term memory LSTM) trained mitigate scale ambiguity thus improve concatenated ego-motion estimation. Compared existing techniques recent effort RNN solutions reconstruction, show our approach competitive accuracy robustness.