Robotics: real-world data, VLA models and diffusion policies enable smoother continuous control

The post states that robot motion has improved rapidly: large real-world datasets and vision-language-action (VLA) models convert visual-language capability into smooth continuous control; diffusion policies can generate coherent action sequences; compliant perception–execution systems reduce jitter.