EDIP–ETLCL: Error-driven incremental policy with event-triggered lifted concurrent learning for hydraulic manipulator trajectory tracking

The high power density and load-bearing capability of hydraulic actuation make hydraulic manipulators well suited to heavy-duty industrial operations. Accurate motion control remains challenging because valve dead zones, hydraulic lag, hysteresis, and friction lead to nonlinear and history-dependent current-to-motion responses. This paper proposes an Error-Driven Incremental Policy with Event-Triggered Lifted Concurrent Learning (EDIP–ETLCL) framework for joint-space trajectory tracking of hydraulic manipulators. The framework combines physics-guided response representation with bounded-memory online adaptation, reducing the dependence on complete hydraulic–mechanical model construction and extensive offline training while retaining a tractable online update structure. ETLCL learns the local current-to-acceleration response from measurable operating signals using hydraulic-specific lifted features, event-triggered sample admission, and concurrent learning, and estimates the corresponding local sensitivity. EDIP uses this sensitivity to convert auxiliary-acceleration regulation errors into valve-current corrections, thereby linking online response learning directly to closed-loop current adaptation. Experiments on a 5-DoF hydraulic manipulator show that ETLCL achieves an average joint-position prediction RMSE of 1.07 × 10 − 2 rad , while the complete EDIP–ETLCL controller achieves an average joint-position tracking RMSE of 1.36 × 10 − 2 rad . Additional ablation, sensitivity, noise-perturbation, comparative, and execution-time analyses further demonstrate the contributions of the principal components, performance advantages under the tested conditions, tolerance to measurement noise, and real-time feasibility on the experimental platform.