Unsupervised Learning for Physical Interaction through Video Prediction
This addresses the challenge of scaling real-world interaction learning for agents by reducing reliance on labeled data, though it is incremental as it builds on existing video prediction methods.
The paper tackles the problem of learning physical object motion without labeled data by developing an action-conditioned video prediction model that predicts pixel motion, enabling generalization to unseen objects. The result is more accurate video predictions, demonstrated on a dataset of 59,000 robot interactions, outperforming prior methods both quantitatively and qualitatively.
A core challenge for an agent learning to interact with the world is to predict how its actions affect objects in its environment. Many existing methods for learning the dynamics of physical interactions require labeled object information. However, to scale real-world interaction learning to a variety of scenes and objects, acquiring labeled data becomes increasingly impractical. To learn about physical object motion without labels, we develop an action-conditioned video prediction model that explicitly models pixel motion, by predicting a distribution over pixel motion from previous frames. Because our model explicitly predicts motion, it is partially invariant to object appearance, enabling it to generalize to previously unseen objects. To explore video prediction for real-world interactive agents, we also introduce a dataset of 59,000 robot interactions involving pushing motions, including a test set with novel objects. In this dataset, accurate prediction of videos conditioned on the robot's future actions amounts to learning a "visual imagination" of different futures based on different courses of action. Our experiments show that our proposed method produces more accurate video predictions both quantitatively and qualitatively, when compared to prior methods.