|
چکیده
|
Many real-world control problems can be modeled as single-agent or multi-agent systems where the agents learn some appropriate behaviors to reach a given goal. To this end, agents are Reinforcement Learning (RL) learners with plenty of different structures and algorithms. RL is an intelligent framework of machine learning concerned with how agents learn optimal policies by interacting with an environment to maximize a reinforce signal, reward. The environment contains whatever is outside of the learner agent and represents the problem to be solved. Using RL algorithms, agents do not need to know the desired system outputs nor the hydraulic, kinematic or dynamic models of the environment. In this chapter, RL techniques are used in combination with Fuzzy Inference Systems (FISs) to design powerful controllers for various tasks. This structure is known as Fuzzy Reinforcement Learning (FRL) and it is often used to create a controller with critic-only architecture. RL algorithms with this architecture have a high degree of exploration and are more easily interpretable due to the embedded expert knowledge in the FISs
|