Prismatic World Model: Learning Compositional Dynamics for Planning in Hybrid Systems

arXiv:2512.08411v2 Announce Type: replace Abstract: Model-based planning in robotic domains is challenged by the hybrid nature of physical dynamics, where continuous motion is punctuated by discrete events such as contacts and impacts. Conventional latent world models typically employ monolithic neural networks that enforce global continuity, which over-smooths distinct dynamic modes (e.g., sticking vs. sliding, flight vs. stance). For a planner, this smoothing results in compounding errors duri