Track manipulation, locomotion and embodied systems.
Search papers, labs, and topics across Lattice.
We introduce PhysicsLENS, a dataset and benchmark designed to diagnose "physical property blindness" in video generation models by evaluating their adherence to underlying physical principles, not just visual fidelity. By constructing matched scenario pairs that vary only the described physical properties (e.g., weight, friction) while keeping conditioning frames and tasks identical, we reveal that current models frequently generate visually plausible videos that violate stated physics. Our evaluation across four models and over 400 human-annotated labels indicates that a significant portion of generated videos (34/47) ignore the specified physical property, highlighting a critical gap in their physical grounding.
Visually plausible robot videos often disregard explicitly stated physical properties like weight and friction, undermining their utility for real-world applications.
Reliable video world models could provide scalable predictive environments for robot learning, planning, and evaluation. However, generated robot videos can violate physical principles and complete tasks through physically implausible behavior, limiting their reliability for robot learning and planning. Current video-generation benchmarks exclude physics that are inherently hidden by visuals (e.g., weight, viscosity, friction). Due to this, video models are evaluated on the fidelity of physics, not the underlying accuracy of physics. We introduce PhysicsLENS, a dataset and benchmark for evaluating plausibility of physical properties grounded in robotics. PhysicsLENS uses matched scenario pairs that hold the same conditioning frame and task, while varying underlying physics in the scene description. Scenarios are curated from public robot video sources and annotated across seven physical domains: collision, gravity, momentum, friction, deformation, fluid, and causality. We evaluate across four video generation models, producing over 400 human-annotated labels. Results show that plausible-looking videos often ignore the stated property (34 of 47), and that stating the property lowers plausibility only slightly and not significantly.