Navigation method for autonomous mobile robots based on ROS and multi-robot improved Q-learning

Recently, path planning of multi-autonomous mobile robot systems is one of the interesting topics in scientific research due to its complexity and its wide use in many fields, such as modern industry, war field, and logistics. Q-learning algorithm which is a sort of reinforcement learning is a widely used method in autonomous mobile robot path planning, thanks to its capacity of learning by itself in any environment without the need for prior knowledge. To increase the convergence speed of the Q-learning algorithm and adapt it to robotics and multi-robot systems, the Multi-Robot Improved Q-Learning algorithm (MRIQL) is proposed. The Artificial Potential Field algorithm (APF) is used to initialize the Q-learning. During learning, a restricting mechanism is used to prevent unnecessary actions while exploring. This Improved Q-learning algorithm is adapted to multi-robot system path planning by controlling and adjusting the policies of the robots to generate an optimal and collision-free path for each robot. We introduce a simulation environment for mobile robots based on Robot Operating System (ROS) and Gazebo. The experimental results and the simulation demonstrate the validity and the efficiency of the proposed algorithm.