The recent advancements in the field of robotics and reinforcement learning have significantly pushed the boundaries of what is possible with humanoid and quadruped robots. The focus has shifted towards developing more robust, adaptive, and versatile control systems that can handle complex and dynamic environments. Key areas of innovation include the integration of deep reinforcement learning with safety constraints, the development of adaptive control strategies for bipedal locomotion on varying terrains, and the use of advanced state estimation techniques to enhance agility and performance. Notably, there is a growing trend towards the use of imitation learning and diffusion models to improve the training efficiency and performance of robotic systems. These approaches aim to bypass the complexities of traditional reinforcement learning by leveraging expert demonstrations and simplifying the training process through non-adversarial methods. Additionally, the field is witnessing a rise in the application of Transformer-based architectures for state estimation, which are proving to be highly effective in dynamic and unpredictable scenarios. Overall, the current research is driving towards more intelligent, adaptable, and safe robotic systems that can operate in real-world environments with greater efficiency and reliability.