Path Planning for Mobile Robots using Deep Learning Architectures

In recent years robots are replacing the human workforce in hospitals, defense, agricultural fields, laboratories, etc,. Majority of the robots working in such areas should possess the ability of path planning to quickly navigate without striking on the obstacles. There are many conventional algorithms for path generation for a given grid map such as A*, Dijikstra, RRT, etc., but they demand more memory resources and take a lot of time. These disadvantages can be overcome by using Deep Learning architectures. In this paper different deep learning architectures like Multi Layer Perceptron, CNN, LSTM, CNN-LSTM for path planning on a grid map are designed and their performances are compared. Among all the deep learning architectures implemented for path planning, CNN-LSTM model performed better with minimum validation Mean Absolute Error(MAE) of 0.0366, and with test MAE of 0.0352, it performed well for dynamic obstacle and moving goal point environments too.