MolmoMotion: Forecasting Point Trajectories in 3D with Language Instruction
MolmoMotion predicts future 3D trajectories of object-attached points in a 3D world using a short video history and a natural language goal description. It uses two prediction methods—autoregressive text decoding and flow-matching—to forecast motion. This method significantly outperforms existing motion prediction baselines on PointMotionBench, providing a general, class-agnostic representation of object movement that transfers well to robotics and video generation tasks.