Training Verifiably Robust Agents Using Set-Based Reinforcement Learning
This work addresses the need for robust reinforcement learning agents in safety-critical applications, though it builds incrementally on existing verification methods.
The paper tackles the problem of neural networks in reinforcement learning being sensitive to input perturbations, which hinders deployment in safety-critical environments, by training agents using entire sets of perturbed inputs to maximize worst-case reward, resulting in verifiably more robust agents than related work as demonstrated on four benchmarks.
Reinforcement learning often uses neural networks to solve complex control tasks. However, neural networks are sensitive to input perturbations, which makes their deployment in safety-critical environments challenging. This work lifts recent results from formally verifying neural networks against such disturbances to reinforcement learning in continuous state and action spaces using reachability analysis. While previous work mainly focuses on adversarial attacks for robust reinforcement learning, we train neural networks utilizing entire sets of perturbed inputs and maximize the worst-case reward. The obtained agents are verifiably more robust than agents obtained by related work, making them more applicable in safety-critical environments. This is demonstrated with an extensive empirical evaluation of four different benchmarks.