FlowMaps: Modeling Long-Term Multimodal Object Dynamics with Flow Matching

FlowMaps is a newly introduced latent flow matching model designed to estimate multimodal distributions of dynamic object locations in 3D spaces, leveraging human interaction patterns for improved navigation. It learns implicit dependencies among objects and their temporal changes, enabling generalization in unseen environments. In extensive testing across over 600 episodes, FlowMaps demonstrated superior performance in dynamic object navigation tasks compared to existing state-of-the-art methods, highlighting its potential for enhancing robotic capabilities in evolving household settings.