In most real world applications, mobile robots should perform their tasks in previously unknown environments. Thus, a mobile robot architecture capable of adaptation is very suitable. This work presents an adaptive architecture for mobile robots, AAREACT, which has the ability of learning how to coordinate primitive behaviors codified by the Potential Fields method by using Reinforcement Learning. The proposed architecture’s performance is compared to that showed by an architecture that performs a fixed coordination of its behaviors, and shows a better performance for different environments. Experiments were performed in the robot Pioneer’s simulator, from ActivMedia Robotics®. The obtained results also suggest that AAREACT has good adaptation skills for specific environment and task.
Mobile robotics; control architecture; reactive behaviors; reinforcement learning
