Modeling the environment of a mobile robot using a time series convolutional neural networks
This thesis present a method for 2D laser scan matching using a 1D time series Convolutional Neural Network to solve SLAM problem. Scan matching is the problem of finding the relative position from two consecutive scans. The algorithm is verified with gazebo simulation data. In addition, a Lucas-Kanade optical flow is implemented to distinguish moving and not moving object. A ROS based system is implemented to integrate deep learning with SLAM and optical flow.
B02974 | (Rack Thesis) | Available |
No other version available