Data-driven NMPC of grading operations for excavators: Approaches and experimental results

Hydraulic excavators are widely used in construction tasks such as levelling and grading, which require high precision and repeatability. Automating these operations is challenging due to the strong nonlinearities, coupling effects, and variability of hydraulic systems. In this work, we present a data-driven Nonlinear Model Predictive Control (NMPC) framework to autonomous grading operations of hydraulic excavators. The system dynamics are modelled using Local Linear Neuro-Fuzzy Models identified from experimental input–output data, enabling a flexible representation without requiring detailed physical modelling. Two NMPC formulations are developed and compared: a trajectory-tracking approach and a path-following approach, the latter allowing for adaptive timing along the desired path. The proposed methods are experimentally validated on a full-scale JCB Hydradig 110 W excavator performing levelling and sloping tasks. The results show that both NMPC approaches achieve accurate tracking performance under realistic operating conditions. Moreover, the path-following formulation demonstrates improved robustness in scenarios where flexibility in execution speed is beneficial. A comparison with a baseline data-driven controller highlights the effectiveness of the proposed approach.