Multi-agent Q-learning Based Navigation in an Unknown Environment

Collaborative task execution in an unknown and dynamic environment is an important and challenging research area in autonomous robotic systems. It is essential to start the task execution in situations like search and rescue at the earliest. However, the time duration between team announcement and the arrival of team members at the location of a task delays the execution of the task. The distributed approaches for task execution assume that the path is known. However, in an environment, say, a building, the position of the doors may not be known, and some of the doors may get closed during task execution. Hence, an agent should first learn the map of the environment. The learning of the map of an unknown environment can be accelerated with multiple agents. This paper proposes a distributed multi-agent Q-learning-based approach for navigation in an unknown environment. The proposed approach is implemented using ARGoS, a realistic multi-robot simulator.