Fig. 1
From: Adversarial attack and defense in reinforcement learning-from AI security view
Examples for adversarial attacks on reinforcement learning. As shown in the first line are the examples for adversarial attack in the field of Atari game. The first image denotes the original clean game background, while the others show the perturbed game background which can be called as “adversarial example”. Huang et al. (2017) found that the adversarial examples which are invisible to human have a significant impact on the game result. Moreover, the second line shows the examples for adversarial attack in the domain of automatic path planning. Same as the first row, the first image represents the original pathfinding map, and the remaining two images denote the adversarial examples generated by noise added. Chen et al. (2018b) found that the trained agent could not find its way correctly under such adversarial examples