Planning of Service Mobile Robot Based on Convolutional LSTM Network

Abstract

This paper proposes method that can complete motion planning in one step using convolutional Long Short-Term Memory (LSTM) network. Service mobile robot movement from the starting position to the target position includes three main tasks: mapping, positioning, motion planning. The convolutional LSTM network mainly is used to complete motion planning. The input of the network is a GRB picture with obstacles, target position, starting position. The outputs of the network are linear velocity and angular velocity of service mobile robot. The convolution layer of the network is to mark obstacles, target position and starting position. LSTM layer describes the time characteristics of movement and full connected layer is used to smoothly fit linear velocity and angular velocity of service mobile robot. The convolutional LSTM network can complete tasks of path finding and control, that is, mapping pictures with obstacles, target position, starting position to linear velocity and angular velocity of service mobile robot. Compared with traditional separate solutions for motion tasks, this method has obvious advantages such as good fault tolerance and complete motion tasks planning in real time. In experiment, a mobile robot “Turtlebot” based on ROS systems was used to verify the effectiveness and convenience of the method for motion planning.