Imply improves object physics consistency in world model predictions
IMPLY: Physically Anchored Consistency for World-Model Rollouts
RoboticsArtificial IntelligenceComputer Vision and Pattern RecognitionMachine Learning
Summary
World models try to guess what happens when an object is pushed at different speeds, but they often produce many possible futures that don’t always agree about the object’s real physical properties like mass and friction. The authors show that just checking if these futures agree with each other isn’t enough to know if the model really understands the object. They introduce IMPLY, a method that uses real physical evidence from known pushes to anchor these predictions, making it easier to tell if a model truly tracks the object or just guesses. This anchored consistency helps pick better predictions and avoid models that give consistent but wrong answers.
What this means in practice
- •For robotics engineers: Improve prediction quality of robot world models by anchoring physical property checks to real calibration pushes, enabling better object tracking and interaction.
- •For autonomous vehicle teams: Enhance scene understanding models by using physically anchored rollouts to verify if object properties are correctly inferred during dynamic scenarios.
Authors
Aman Mehta, Riya Baviskar
Abstract
A world model asked what happens if an object is pushed at several speeds produces several futures. If the model has the object in mind, those futures agree about it: each implies the same mass and friction. The consistency checks now used to vet world-action models ask whether a model's futures agree with each other, and none of them knows any physics. We show that this is not enough, and what to do instead. IMPLY reads the physics each rollout implies by inverting a simulator and scores a set of rollouts by how well one object explains all of them, anchored to two calibration pushes the model has observed. In a controlled setting, self-consistency gives a perfect score to a model that ignores the object and always predicts a typical push; anchoring exposes it (AUROC 0.70 versus 1.00). On a real model, V-JEPA 2-AC adapted to the scene, the same thing happens. Given its own calibration pushes the model tracks the object (per-object correlation with the truth 0.91); given another object's, it does not (0.05). Self-consistency cannot tell these apart, preferring the right evidence on 52% of objects, chance level, while anchored disagreement prefers it on 73% and correlates 0.92-0.99 with the rollouts' error. Used to choose among candidate rollout sets, it comes within 0.003 of an oracle that sees the truth. A model that has internalised the wrong object is exactly as self-consistent as one that has internalised the right one; consistency has to be anchored to evidence.