World-model repairs vary with shared or isolated input errors
Same World, Different Knowledge: When Isolated Audits Misjudge World-Model Repairs
Robotics
Summary
When parts of a system get faulty information, fixing one part alone may not help the whole system as expected. The authors look at when missing or uncertain data in models affect predictions and control, especially in drones. They find that some fixes work well when errors happen in isolation but fail when errors affect multiple parts together. Their tests show repairing models needs to consider how information flows between parts to pick the best solution.
What this means in practice
- •For robotic control engineers: Optimize repairs on control models by considering how shared sensor errors affect overall system predictions and performance.
- •For drone system developers: Improve drone flight models by using uncertainty training to recover performance lost due to calibration mistakes shared across modules.
Tested on simulated data.
Authors
Rui Min, Xianyao Li, Fang Xu, Sofiane Lachab, Jing Du
Abstract
A repair favored under an isolated input fault can be inferior when deployed modules share the faulty information. We introduce an information-interface audit for world models, distinguishing fidelity gaps, where exact inputs become estimates, from availability gaps, where inputs are missing. Fixed-weight interventions measure prediction error, input dependence, and paired closed-loop benefit, including dependencies introduced by reconstruction. In simulated quadrotor model predictive control, coupled, opposite-sign 10% mass/thrust calibration errors reduce a physics-anchored model's success from 69% to 8%; uncertainty training restores 65%. Wind reconstruction recovers control benefit but inherits calibration dependence. For a positive calibration offset, reconstruction-only corruption favors uncertainty-trained reconstruction, whereas shared corruption favors the baseline. Acceleration diagnostics reveal compensation between reconstruction bias and nominal-model error, also observed with a disturbance observer. Repair selection therefore depends on the information paths used in deployment.