TMT: Runtime Backdoor Detection for Vision-Language-Action Policies on Unseen Tasks

Key points

• Detecting such activation is difficult because malicious behavior can comprise individually plausible actions, while unfamiliar tasks introduce legitimate changes in observations and behavior.1
• We introduce TMT, a runtime backdoor detector based on Token Manifold and latent Transition modeling.1
• We further explore policy purification through self-distillation: a frozen copy of the backdoored policy provides benign-input actions to supervise a student on paired benign and triggered observations, without requiring a separate clean reference policy.1
• In a post-hoc comparison with ten baselines, TMT achieves state-of-the-art backdoor detection performance on unseen tasks across three VLA backdoor attacks.1

Sources (1)

• [1]TMT: Runtime Backdoor Detection for Vision-Language-Action Policies on Unseen Tasks
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 05:20 AM

  Backdoored vision-language-action (VLA) policies can preserve benign task performance while producing malicious actions when a trigger appears.

  Detecting such activation is difficult because malicious behavior can comprise individually plausible actions, while unfamiliar tasks introduce legitimate changes in observations and behavior.

Extractive summary: sentences quoted from the sources.