Modeling, parameter identification and optimal control of an autonomous novel agricultural machine

The dissertation addresses challenges in the development of autonomous systems for
agricultural applications, using a novel tractor as its experimental
platform. Our contributions span three core areas: model-based parameter
identification, the realization of a Digital Twin, and the design of an
optimal path-tracking controller.
Compared with the attention given to conventional road vehicle dynamics,
off-road agricultural vehicle dynamics remains a relatively unexplored topic. A
central challenge in this field lies in the identification of the parameters that
govern the contact interaction between wheels and soil, as it critically influences
the vehicle’s behavior. This means that, when developing a mathematical model
describing the tractor dynamics, precise knowledge of these parameters is required
in order to obtain a model capable of accurately predicting the system’s
behavior. Traditional approaches to parameter identification typically rely on extensive
sensor instrumentation and controlled, lab-based experiments, rendering
them impractical for daily field operations. In this work, we propose an approach
that combines standard operational data, obtained from satellite and on-board
systems, with mathematical models of the vehicle behavior, for the purpose of
identifying system parameters. This allows for model-based identification without
the need for additional measurement equipment. The resulting parametric models
are then evaluated using performance metrics assessing their predictive accuracy.
Based on these models, we propose a Digital Twin architecture leveraging
state-of-the-art simulation tools to provide a digital representation of the system,
enabling virtual testing, and a consequent seamless transition from simulation to
deployment in development processes.
The utility of a Digital Twin is demonstrated through the development of a
Model Predictive Control (MPC) strategy for accurate path tracking. High positional
accuracy in path tracking is indeed required in advanced agricultural
approaches, such as the so-called precision agriculture framework, in order to
actually achieve the targeted efÏciency gains. The proposed controller, which is
based on a simplified model of the kinematic behavior of the tractor, is first validated
in simulation, and then successfully deployed in field experiments, showcasing
the Digital Twin as a key enabler for bridging the gap between control
design and real-world implementation.