Robots Taught to Expect the Worst: Adversarial Fine-Tuning Makes AI Control Systems Fail-Safe
Researchers have developed an offline-to-online reinforcement learning framework that uses adversarial action perturbations and an adaptive curriculum to make robot ...
