Researchers have developed DECOWAM, a new model for legged mobile robots that can predict how their movements affect future observations. This is achieved by separating the robot's camera motion from its base and arm actions. The model was tested on a real-robot dataset and showed improved performance in predicting both video and action outcomes compared to existing models. --- Why it matters: This matters because it enables more accurate prediction of a robot's interactions with its environment, which is crucial for tasks like manipulation and navigation. By separating the factors that influence these interactions, DECOWAM can improve whole-body coordination and control under changing viewpoints. Source: https://arxiv.org/abs/2608.20114