Search papers, labs, and topics across Lattice.
This paper introduces a simulation-trained variable impedance control framework for multitask exoskeletons, addressing the challenge of ensuring interaction safety while adapting to diverse activities. By employing Proximal Policy Optimization (PPO) to generate human muscle activation data and training a dual modality policy that integrates semantic instructions with proprioceptive history, the authors achieve effective prediction of reference trajectories and impedance gains. Experimental results demonstrate that this approach significantly reduces metabolic costs in real-world applications compared to standard methods, highlighting its potential for safe and efficient exoskeleton control.
A novel variable impedance control framework for exoskeletons not only enhances task adaptability but also ensures safety through stability constraints derived from Lyapunov theory.
Wearable exoskeletons can augment human phys ical capabilities during complex activities. However, ensuring adaptation across diverse tasks while guaranteeing interaction safety remains a critical challenge. To address this, a simulation trained variable impedance control approach with stability guarantees is proposed. First, a simulation-based human exoskeleton motion data generation pipeline is established, utilizing Proximal Policy Optimization (PPO) to synthesize human muscle activations while the exoskeleton provides direct compensation for human biological joint torques. Subsequently, the generated dataset is used to train a dual modality policy that fuses semantic instructions with proprioceptive history, enabling the prediction of reference trajectories and variable impedance gains for nine different motion tasks. To guarantee safety, the network outputs are constrained by a stability criterion derived from Lyapunov stability theory, which bounds stiffness variations to ensure the asymptotic stability of the coupled system. Experimental results indicate that the proposed framework reduces metabolic cost in real-world scenarios com pared with standard baseline methods. These findings suggest the feasibility of the proposed framework for safe, multitask exoskeleton control.
