BadWAM: When World-Action Models Dream Right but Act Wrong
Machine LearningRobotics
Summary
The authors study World-Action Models (WAMs), which link a robot's actions to its predictions about the future world to help control tasks better. They show that this connection can be tricked by a new kind of attack called World-Action Drift Attacks, which use tiny changes in the robot's visual input to confuse it. These attacks can either force clearly wrong actions or secretly cause mismatches where the robot imagines a reasonable future but actually acts incorrectly. Their experiments demonstrate that these attacks significantly lower the robot's success in tasks, revealing weaknesses in how WAMs predict and act.
Authors
Qi Li, Xingyi Yang, Xinchao Wang
Abstract
World-action models (WAMs) are emerging as a promising foundation for embodied control: rather than predicting actions alone, they learn representations that couple action generation with future world prediction. This coupling is often viewed as a source of robustness, interpretability, and safety, as a robot's action can in principle be checked against its imagined future. In this paper, we show that this assumption is fragile. We introduce BadWAM, a unified framework for modeling and evaluating World-Action Drift Attacks: a new class of WAM-specific adversarial attacks that use small visual perturbations to break the alignment between what a WAM imagines and what it executes. BadWAM characterizes this attack surface along two natural criteria: attack strength and stealthiness. When the adversary prioritizes disruption, BadWAM instantiates an action-only adversarial attack, which directly drives the model toward task-failing actions. When the adversary additionally prioritizes stealth, BadWAM instantiates an imagination-preserving adversarial attack, which seeks to induce harmful action shifts while keeping the model's predicted future close to its clean imagination. Together, these two attacks capture a spectrum of WAM-specific failures: from overt action hijacking to stealthier cases where the model appears to imagine a plausible future but executes a desynchronized action. We evaluate BadWAM across different variants of WAMs. Results show that our attacks substantially reduce task success rates under closed-loop execution. For example, our action-only attack reduces the model performance from 96.5% to 43.1% success. The results of our imagination-preserving attack further exposes a WAM-specific vulnerability: moderate future-preserving regularization can maintain strong attack performance while reducing future imagination drift.