Reinforcement Learning Based Autonomous Path Finding Robot for Dynamic Environments

  • Rajasekaran Thangaraj, Sivaramakrishnan Rajendar, Premkumar Murugiah, Vidhya Kandasamy, Rahul R


In real world the path identification for autonomous robot in dynamic environment is difficult process.
However, this paper introduce a reinforcement learning based double DQN learning model for
autonomous robot to handle the problem of dynamic path identification. The experiment is performed to
evaluate the effectiveness of the algorithm using the simulation tool Udacity. The fundamental idea is to
determine the problem as state-action problem and discover the value for each state in its workspace. The
performance of the model is examined in two different scenarios such as normal path and hill path. The
results of the experiment confirm that the model shows better performance in normal path with an
accuracy of 92% than hill path which is of 86%. Simulation results confirms that the reinforcement
learning based model for autonomous robot provide better accuracy in dynamic environment for normal
path than the hill path