Schrödinger's Cat: Probabilistic Representation and Prediction of Potential Scene Kinematics
GARFIELD models a distribution of possible scene motions instead of betting on one future.
The paper describes a probabilistic kinematics model that works from an image and optional sparse spatio-temporal constraints. Its latent representation can sample joint trajectories and decode motion density directly, letting uncertainty be tied to particular objects and timesteps. The authors report motion-planning results competitive with large video generation models, with trajectory sampling 97x faster. They also claim motion-density estimates run two orders of magnitude faster than Monte Carlo sampling from motion generation models. ArXiv · AI/CL/LG's note
The paper describes a probabilistic kinematics model that works from an image and optional sparse spatio-temporal constraints. Its latent representation can sample joint trajectories and decode motion density directly, letting uncertainty be tied to particular objects and timesteps. The authors report motion-planning results competitive with large video generation models, with trajectory sampling 97x faster. They also claim motion-density estimates run two orders of magnitude faster than Monte Carlo sampling from motion generation models. ArXiv · AI/CL/LG's note
score 4