Optimizing human-in-the-loop training: Real-time personalized task scheduling via model predictive control

Effective motor learning is essential across domains such as rehabilitation and robot-assisted professional training, with task scheduling playing a key role when learning multiple tasks. Prior studies have shown that, when using the dual-rate motor learning model, task learning involves a slow-learning, slow-forgetting state, and the schedules that maximize this state during learning enhance long-term retention. However, existing task scheduling methods are either rule-based and not grounded in motor learning models, limiting their optimality, or rely on offline optimization, restricting their real-time applicability. In this study, we introduce a novel control-theory-based framework that integrates Moving Horizon Estimation (MHE) and Model Predictive Control (MPC) to enable real-time optimization of task schedules in human-in-the-loop training. Earlier work demonstrated that MHE can reliably estimate the states and parameters of the dual-rate model. In the present work, we leverage these real-time estimates within an MPC framework to select the optimal task sequence that maximizes the slow-learning state, thereby promoting long-term retention. In simulations, MPC was benchmarked against an Alternating schedule and offline optimization methods, consistently outperforming the Alternating schedule by up to a 27% lower cost function value and closely matching offline optimization. In the experiment (n = 20; Alternating vs. MPC), baseline performance was comparable, but the MPC group showed mean motor errors up to 46% lower across later testing phases, indicating significantly faster adaptation and stronger retention (p < 0.05). These findings demonstrate the feasibility of real-time, personalized training scheduling and highlight its potential to transform next-generation rehabilitation and high-performance skill acquisition.