<ul data-eligibleForWebStory="true"><li>Deep reinforcement learning (DRL) has been successful in various domains, but it is susceptible to backdoor attacks during training.</li><li>Existing backdoor attacks in DRL often use simplistic trigger configurations.</li><li>A new framework called TooBadRL focuses on optimizing DRL backdoor triggers in terms of timing, spatial dimensions, and magnitude.</li><li>TooBadRL introduces an adaptive freezing mechanism for injection timing and a cooperative game approach to select influential state variables.</li><li>Furthermore, TooBadRL utilizes a gradient-based adversarial method to optimize injection magnitude under environment constraints.</li><li>Evaluation on three DRL algorithms and nine tasks shows that TooBadRL increases attack success rates while maintaining normal task performance.</li><li>This research emphasizes the significance of systematic trigger optimization in DRL backdoor attacks.</li><li>The TooBadRL framework's source code is available at https://github.com/S3IC-Lab/TooBadRL.</li></ul>

TooBadRL: Trigger Optimization to Boost Effectiveness of Backdoor Attacks on Deep Reinforcement Learning

Discover more