Learning-based motion predictor for human–robot-interactive system with large communication delays

This paper investigates the motion prediction problem for human–robot-interactive (HRI) system subject to large communication delays. A data-driven inverse optimal control (IOC) approach is applied to identify the human input through learning from the trajectory demonstrations of the robot end-effector. Based on the identification results, a motion predictor with cascade structure is proposed to estimate the real-time states of HRI system using the delayed discrete measurements, which consists of a continuous-discrete time observer and several sub-predictors in a chain. The proposed predictor can handle with arbitrary large delays through utilizing sufficient numbers of sub-predictors. The convergence of prediction errors is demonstrated via Lyapunov–Krasovskii method. Simulations of a multi-multi-degree-of-freedom HRI system verifies the ability of proposed predictor to perform an accurate estimation under large delays and noises.