Food production in agriculture is facing significant challenges from population growth and labour shortages, increasing the need to automate labour-intensive tasks. Omnidirectional mobile robots (OMRs) with three planar degrees of freedom are well suited to navigating narrow greenhouse aisles for harvesting, pruning, pest detection, and related applications. This thesis identifies a nonlinear model of such an OMR and designs a trajectory-tracking controller for the MIRTE Master platform developed in the Cognitive Robotics department at TU Delft. Limited data availability and onboard computation make this problem especially challenging.

Sparse identification of nonlinear dynamics with control (SINDYc) is used to learn a sparse, physics-informed model of the MIRTE Master from data. On top of this model, robust tube-based nonlinear model predictive control (NMPC) is implemented for real-time trajectory tracking. The controller uses a two-layer structure that combines nominal planning with an invariant error tube to reject disturbances and model mismatch.

The proposed modelling and control pipeline is validated both in simulation and experimentally on the robot. The results demonstrate real-time feasibility and robust tracking performance, supporting the development of reliable and efficient control systems for agricultural OMRs.