Virtual Impedance Adaptation of Lower-Limb Exoskeleton for Human Performance Augmentation Based on Deep Reinforcement Learning

• Proposes a novel virtual impedance adaptation strategy (VIADRL) for lower-limb exoskeletons, leveraging deep reinforcement learning to reduce reliance on accurate dynamic models. • Introduces a hybrid inverse-forward dynamics simulation approach for safe and efficient policy training in a multibody environment. • Demonstrates significant reduction in human-exoskeleton interaction forces (up to 19% in simulation and 11% on prototype) compared to the sensitivity amplification control with deep reinforcement learning (SADRL) benchmark. • Provides a quantitative evaluation method based on HEI forces at back, thighs, and shanks, offering a new metric for exoskeleton control performance.