The recent advancements in robotic manipulation and locomotion have seen a significant shift towards leveraging diffusion models and probabilistic approaches to enhance policy learning and generalization. A notable trend is the integration of diffusion processes into various stages of robotic task execution, from policy generation to motion optimization. This approach allows for more robust and diverse policy distributions, which are crucial for handling complex, multimodal tasks in dynamic environments. Additionally, there is a growing emphasis on data-efficient learning, with methods like Latent Weight Diffusion and Causal Attention Enabling Data-Efficient Generalizable Robotic Manipulation (CAGE) demonstrating substantial improvements in generalization with limited demonstrations. These innovations not only reduce the computational burden but also enhance the adaptability of robotic systems to new tasks and environments. Furthermore, the use of probabilistic models for skill acquisition and subgoal mapping is advancing the field by enabling robots to learn from heterogeneous action spaces and improve their robustness to observation noise. Notably, DiffusionSeeder stands out for its ability to seed motion optimization with diffusion, significantly speeding up planning in complex environments while maintaining high success rates. Overall, the field is progressing towards more efficient, robust, and versatile robotic systems capable of handling a wide range of tasks with minimal human intervention.