Swarmand pheromone based reinforcement learning methods for the robot(s) path search problem
Neha Khanduja, Nupur Jha, Bharat Bhushan · 2016
With the world moving to an automated platform, robots are finding application in almost all domains to reduce the human effort. One such domain is to find a path in an unknown and hostile environment to reach the goal. Due to the complexity of many tasks in this domain itis difficult for the robots (agents) to solve this with pre-programmed agent behaviors. Instead of using pre-programmed agent behaviors agents must discover a solution on their own by using learning. In simple reinforcement learning algorithms, a single agent learns how to achieve a goal through many episodes. But if complexity of learning problem is increased or the number of agents is more, it may lead to take more computation time in order to obtain the optimal policy and sometimes it may be the situation that it may not reach to the goal. For optimization problems, bio inspired multi-agent search methods such as particle swarm optimization, ant colony optimization have been recognized to obtain a global optimal solution for multi-modal functions with wide solution space rapidly. This paper proposes a SARSA based reinforcement learning algorithm using one agent and two agents where the agents are guided by the pheromone levels also called the Phe-SARSA. In this algorithm agents learn through not only their respective experiences but also with the help of pheromone trail left by other agents to search for the shortest path. The algorithms have been simulated in the MATLAB 2013a and the results have been compared with the Q-learning, SARSA, Q-Swarm, SARSA-Swarm and Phe-Q algorithms.