AI RESEARCH
Kinema4D: Kinematic 4D World Modeling for Spatiotemporal Embodied Simulation
arXiv CS.CV
•
ArXi:2603.16669v1 Announce Type: cross Simulating robot-world interactions is a cornerstone of Embodied AI. Recently, a few works have shown promise in leveraging video generations to transcend the rigid visual/physical constraints of traditional simulators. However, they primarily operate in 2D space or are guided by static environmental cues, ignoring the fundamental reality that robot-world interactions are inherently 4D spatiotemporal events that require precise interactive modeling. To re this 4D essence while ensuring the precise robot control, we