Imagined Rollouts are Kinematic, Not Dynamic: A Diagnosis of Long-Horizon World-Model Failure
For researchers building world models for long-horizon planning, this work provides a concrete diagnostic to distinguish kinematic from dynamic failure modes, though it is an incremental diagnostic rather than a new method.
The paper identifies that long-horizon failures in world models are due to kinematic rather than dynamic imagination, introducing a diagnostic (iKCE) that shows DreamerV3's imagined rollouts have iKCE two orders of magnitude above real-physics rollouts and remain flat across a friction sweep where policy reward collapses.
Long-horizon failure in world models is conventionally attributed to compounding error, a generic framing that does not distinguish what kind of error compounds. We propose a kinematic-vs-dynamic reframing: world models tend to imagine kinematically rather than dynamically. We operationalize this as the imagined Kinematic-Consistency Error, a per-step diagnostic that measures how far a rollout departs from a closed-form kinematic null, paired with a perturbation protocol that tests whether iKCE responds when physical conditions cross a regime boundary. We instantiate the diagnostic on a released DreamerV3 checkpoint trained on DMC walker-walk, where imagined iKCE runs roughly two orders of magnitude above that of matched real-physics rollouts. Across a friction sweep that crosses the gait-collapse boundary, the model's iKCE stays statistically flat even as the trained policy's reward collapses through the same range, providing the kinematic-not-dynamic signature. The diagnostic distinguishes kinematic from dynamic imagination at horizons longer than the embodiment's gait period.