Sequential Attacks on Agents for Long-Term Adversarial Goals
This exposes serious security threats for RL applications in safety-critical systems like drones and self-driving cars, representing a novel method for a known bottleneck.
The paper tackles the vulnerability of reinforcement learning agents with deep neural network policies by showing that a sequence of small adversarial perturbations can impose an arbitrary adversarial reward, misleading the agent to optimize for it over time, using the Adversarial Transformer Network method.
Reinforcement learning (RL) has advanced greatly in the past few years with the employment of effective deep neural networks (DNNs) on the policy networks. With the great effectiveness came serious vulnerability issues with DNNs that small adversarial perturbations on the input can change the output of the network. Several works have pointed out that learned agents with a DNN policy network can be manipulated against achieving the original task through a sequence of small perturbations on the input states. In this paper, we demonstrate furthermore that it is also possible to impose an arbitrary adversarial reward on the victim policy network through a sequence of attacks. Our method involves the latest adversarial attack technique, Adversarial Transformer Network (ATN), that learns to generate the attack and is easy to integrate into the policy network. As a result of our attack, the victim agent is misguided to optimise for the adversarial reward over time. Our results expose serious security threats for RL applications in safety-critical systems including drones, medical analysis, and self-driving cars.