Autonomous robots employed for infrastructure inspection, disaster responses and monitoring involve navigation within an unknown environment with an ambiguous goal.
For safe navigation, the robot should be equipped with mapping and localization module(SLAM) which is capable of exploring in the presence of obstructions using a 360 degree LIDAR. To-date well-known approach is to navigate the robot to the boundary of the explored area. However, recognising the optimal boundaries for faster and energy-efficient exploration is still a challenge considering the partially observable Markov decision process involved in selecting the navigation way-points.
The research comprises developing a novel parameterization of the exploration algorithm capable of energy efficient autonomous navigation ensuring efficient mapping in minimum time. The exploration function employed will be optimized using reinforcement learning to handle environmental stochasticity. The developed algorithm once experimented in gazebo based simulation environment will be deployed on a real robot to analyze the performance.