This survey explores the application of reinforcement learning (RL) in robotics, highlighting both challenges and successes in designing complex behaviors for robots. RL provides a framework for autonomous behavior generation, while robotic problems inspire and validate RL developments. The paper discusses key challenges, such as high-dimensional state and action spaces, partial observability, and the need for efficient learning in real-world environments. It emphasizes the choice between model-based and model-free methods, as well as value function-based and policy search approaches. The paper also addresses the importance of algorithms, representations, and prior knowledge in achieving success in RL for robotics. It highlights the role of reward shaping, the generation of appropriate reward functions, and the challenges of under-modeling and model uncertainty. The paper reviews various RL methods, including value function approaches and policy search, and discusses their applicability in robotics. It also explores the relationship between RL and optimal control, noting that RL can handle complex reward structures and sequential interactions. The paper concludes by emphasizing the potential of RL in robotics and the need for further research in this area.This survey explores the application of reinforcement learning (RL) in robotics, highlighting both challenges and successes in designing complex behaviors for robots. RL provides a framework for autonomous behavior generation, while robotic problems inspire and validate RL developments. The paper discusses key challenges, such as high-dimensional state and action spaces, partial observability, and the need for efficient learning in real-world environments. It emphasizes the choice between model-based and model-free methods, as well as value function-based and policy search approaches. The paper also addresses the importance of algorithms, representations, and prior knowledge in achieving success in RL for robotics. It highlights the role of reward shaping, the generation of appropriate reward functions, and the challenges of under-modeling and model uncertainty. The paper reviews various RL methods, including value function approaches and policy search, and discusses their applicability in robotics. It also explores the relationship between RL and optimal control, noting that RL can handle complex reward structures and sequential interactions. The paper concludes by emphasizing the potential of RL in robotics and the need for further research in this area.