The field of robotics involving autonomous mobile robots has become a significant research area as it offers improved productivity by reducing human intervention. In any such system, path planning is one of the most basic and crucial components wherein the objective is to determine a collision-free path to the goal in an obstacle rich environment. In the last two decades, sampling based methods like Rapidly Exploring Random Tree (RRT*) and Fast Marching Trees(FMT*) have been widely used for solving path planning problems in higher dimensions. However they involve significant computational overhead for finding the path. In recent years, the field of Reinforcement Learning (RL) combined with deep neural networks has enabled tackling decision-making tasks involving complex environments. In this talk, we aim to solve the complex real-time path planning problem using the Deep-RL framework.

The path planning problem consists of finding a path in real time for a robot from the given initial position to the goal position. The environment is assumed to be mapped (known completely) and the resulting path should avoid all the obstacles in the mapped environment. The robot's dynamics is assumed to be non-holonomous with the presence of process noise and is controlled with a finite set of inputs. The environment is stated in the form of a Markov Decision Process framework and a Deep Reinforcement Learning framework is used for solving the problem. Simulations are performed for the proposed method using Deep Q-Network based algorithm where the agent is trained to learn the optimal policy by designing appropriate reward functions. The results obtained are compared with sampling based path planning algorithms like Real Time-RRT* and Robust-FMT* for both static and dynamic environments. It is observed that the computational time of the proposed Deep-RL method is significantly faster for real-time operations compared to the sampling based methods with a marginal drop in optimality. The proposed method is validated by implementing it on a Turtlebot robot (Kobuki).