Back to the Future: Predicting Long-Horizon Dynamics with Retrospective Correction for Disturbance-Aware Retrieval

This research presents DARE (Disturbance-Aware REtrieval), a geometry-aware relational dynamics framework designed to predict how contact disturbs surrounding objects during retrieval from densely packed storage. The method uses retrospective correction to amend earlier predicted states when sparse observations become available, improving long-horizon prediction reliability. Experiments in planar and RoboCasa environments evaluate the framework's performance on long-horizon prediction, generalization, and robustness for disturbance-aware retrieval.