VLA-Forget: Vision-Language-Action Unlearning for Embodied Foundation Models

Researchers introduce VLA-Forget, a new unlearning framework for vision-language-action (VLA) models used in robotic manipulation. The hybrid approach addresses the challenge of removing unsafe or unwanted behaviors from embodied AI foundation models while preserving their core perception, language, and action capabilities.

• →VLA models face unique unlearning challenges as undesirable knowledge can be distributed across perception, alignment, and reasoning layers.
• →VLA-Forget combines ratio-aware selective editing for perception with layer-selective reasoning unlearning to preserve utility.
• →The framework improves forgetting efficacy by 10% while preserving perceptual specificity by 22% compared to baseline methods.
• →Reasoning and task success are retained at 9% higher rates with 55% reduced post-quantization recovery.
• →The approach addresses safety and privacy concerns in deploying embodied AI foundation models for robotics.