SpatioTemporal Information

SpatioTemporal is building Motion Intelligence for Physical AI. Robots can increasingly see what is around them and plan what to do. But they still lack the instinct in between: understanding what movement means, what is likely to happen next, and whether it is safe to act around people. SpatioTemporal is building a family of Spatial and Temporal Intelligences for robotics and autonomous systems. Our first foundation model is Motion Intelligence: a learned vocabulary of motion tokens that compresses space and time into machine-readable movement patterns. Rather than learning from pixels alone, we model how people, vehicles and objects move: hesitation, drift, yielding, acceleration, stopping, turning and intent as it emerges over time. Motion Intelligence is the first step toward a broader missing middle stack between perception and planning: Spatial Intelligences - motion understanding, intent analysis, human awareness and world-state modelling. Temporal Intelligences - future-state prediction, causal reasoning, consequence modelling and planning handoff. In early NVIDIA Cosmos simulations, adding our Motion Intelligence model reduced robot-human near-collisions from 24% to 2%, with robots yielding earlier and moving more smoothly. Our mission is to give machines the instincts needed to operate safely in shared human spaces. Robots that read the room. Cars that read the road.