Application of single agent Q-learning for light exploration

摘要

Machine learning refers to systematic design and development of algorithms that allows computers to evolve behaviors based on some realistic data (online or offline). Q-learning, a sub-part of the reinforcement learning is being used world wide for easy learning of mobile robots. Light exploration is one of the important issues for developing green robots. This paper describes the work carried out for light exploration by a robot using single-agent based Q-learning. Here a single agent is taking care of all the tasks for learning. ARBIB III, an indigenous behaviour-based robot has been used to implement the Q-learning algorithm for light exploration. The system uses one light sensor and two touch (press) sensors for exploration. It has been found that the algorithm has good applicability for robot learning.