バックドアが部分観測性と出会うとき:実世界強化学習への攻撃
When Backdoors Meet Partial Observability: Attacking Real-World Reinforcement Learning
実世界の強化学習ではLiDARやオドメトリなど制御不能な補助状態が混在するため、条件付き拡散で確率的トリガー分布を学習し、視覚パッチによるバックドア攻撃を安定して発動させる手法を提案した。
詳しい要約
1. どんなもの?
2. 先行研究と比べてどこがすごい?
3. 技術・手法の肝は?
4. どうやって有効だと検証した?
5. 議論はある?
6. 次に読むべき論文は?
※ AIが要旨から生成した要約です。正確性は原文をご確認ください。
著者: Tairan Huang, Qingqing Ye, Yulin Jin, Jiawei Lian, Yaxin Xiao, Yi Wang, Haibo Hu
分類: cs.RO, cs.CV
原文アブストラクト
Backdoor attacks can cause reinforcement learning (RL) policies to behave normally under clean inputs while executing malicious behaviors when triggers are present. Existing RL backdoor attacks are primarily studied in simulation and often assume that attackers can reliably manipulate the observations driving policy decisions. This assumption becomes fragile in real-world deployment, where RL policies commonly rely on multimodal observations. Attackers can manipulate visual inputs through physical triggers, but auxiliary states such as LiDAR and odometry signals remain uncontrollable and vary across trajectories. We study this overlooked challenge and propose a diffusion-guided backdoor attack framework (DGBA) for real-world RL. DGBA uses small printable visual patches as triggers and learns a stochastic trigger distribution via conditional diffusion to maintain consistent attack activation under varying uncontrollable states. We further introduce an advantage-based poisoning strategy that injects triggers only at decision-critical training states. Experiments on a physical TurtleBot3 platform show that DGBA consistently outperforms prior RL backdoor attacks while preserving normal task performance. Demo videos and code are available in the supplementary material.
関連論文
- 信頼できないシミュレータに注意:強化学習における報酬なしバックドア攻撃強化学習セキュリティ
