Certifiable Robustness to Adversarial State Uncertainty in Deep Reinforcement Learning
This work addresses safety-critical applications like autonomous vehicles by providing formal robustness guarantees, though it is incremental as it extends prior research with new guarantees and extensions.
The paper tackles the problem of ensuring deep reinforcement learning policies are robust to adversarial perturbations or noise in sensor inputs, by developing an online certifiably robust defense that computes guaranteed lower bounds on state-action values to choose robust actions under worst-case deviations, and demonstrates increased robustness in pedestrian collision avoidance and control tasks.
Deep Neural Network-based systems are now the state-of-the-art in many robotics tasks, but their application in safety-critical domains remains dangerous without formal guarantees on network robustness. Small perturbations to sensor inputs (from noise or adversarial examples) are often enough to change network-based decisions, which was recently shown to cause an autonomous vehicle to swerve into another lane. In light of these dangers, numerous algorithms have been developed as defensive mechanisms from these adversarial inputs, some of which provide formal robustness guarantees or certificates. This work leverages research on certified adversarial robustness to develop an online certifiably robust for deep reinforcement learning algorithms. The proposed defense computes guaranteed lower bounds on state-action values during execution to identify and choose a robust action under a worst-case deviation in input space due to possible adversaries or noise. Moreover, the resulting policy comes with a certificate of solution quality, even though the true state and optimal action are unknown to the certifier due to the perturbations. The approach is demonstrated on a Deep Q-Network policy and is shown to increase robustness to noise and adversaries in pedestrian collision avoidance scenarios and a classic control task. This work extends one of our prior works with new performance guarantees, extensions to other RL algorithms, expanded results aggregated across more scenarios, an extension into scenarios with adversarial behavior, comparisons with a more computationally expensive method, and visualizations that provide intuition about the robustness algorithm.