Humanoid robots have long been limited in their ability to autonomously perform tasks in diverse environments. This limitation primarily stemmed from challenges in developing generalizable skills. The introduction of the Improved 3D Diffusion Policy (iDP3) marks a significant breakthrough in this area, allowing robots to operate autonomously in varied real-world scenarios. Unlike previous models, iDP3 utilizes egocentric 3D visual representations, circumventing the need for precise camera calibration and segmentation. This innovation permits adaptable humanoid robots to efficiently learn and execute skills using data predominantly collected from controlled environments. The iDP3 system is built upon four critical components: a humanoid robot platform, a sophisticated data collection system, an advanced learning methodology, and real-world deployment. Utilizing the Apple Vision Pro for teleoperation, the system captures precise human movements which the robot mimics through Relaxed IK. This allows the robot to adapt well to changes in its environment and perform tasks such as grasping, even when viewed from different angles. The novel egocentric approach not only provides flexibility but also enables the robot to generalize tasks to a variety of untrained objects and settings. Despite the promising enhancements introduced with iDP3, several challenges remain. The current data collection process, while intuitive, is physically demanding over prolonged periods, and the reliance on less reliable depth sensing technology suggests room for improvement. The nuances of fine motor skills still represent a hurdle, particularly in teleoperation through devices like Apple Vision Pro. Nonetheless, the ability of iDP3 to generalize across diverse environments presents a substantial step forward in humanoid robotics, hinting at future advances where such systems can execute complex tasks autonomously in dynamic and unpredictable real-world conditions.