FlowMaps: Modeling Long-Term Multimodal Object Dynamics with Flow Matching
Researchers have introduced FlowMaps, a novel latent flow matching model designed to predict the future locations of dynamic objects in 3D spaces. This model learns implicit dependencies among objects and their temporal evolution, enabling robots to understand and exploit human routines for tasks like navigation. FlowMaps has demonstrated superior performance in dynamic object navigation tasks across simulated and real-world environments, outperforming existing state-of-the-art methods. AI
IMPACT Enhances robot navigation capabilities by modeling object dynamics and human routines.