From Italy to the Netherlands to advance soft-robot modelling

From May to September 2025, Michele Pierallini, researcher at the University of Pisa (Italy), completed his secondment at the Robotics and Mechatronics (RAM) Laboratory of the University of Twente (The Netherlands). His work focused on developing learned and physics-informed models for underactuated soft robotic systems: an approach that combines physical principles with real-world data to better predict complex robot dynamics.

During his stay, Michele collaborated with the RAM team to investigate physically informed neural networks for modelling an elastic cable attached to a drone. He implemented and tested three architectures (feedforward, Lagrangian, and Transformer neural networks), showing that the physics-informed solutions outperform standard approaches in predicting cable deformation during drone motion. Ten experimental trials were carried out to evaluate accuracy and real-time performance, with all models running five times faster than the drone controller and achieving tracking errors below 5 cm.

This research directly contributes to the objectives of the NeutraWeed project by enabling more accurate and adaptive control of aerial robotic systems for precision weed management. In particular, the developed models support stable and safe operation of drones equipped with flexible tools or powering tethers, which has potential to aid localized weed removal in close proximity to crops. The ability to predict and control flexible elements in real time enhances interaction reliability in unstructured agricultural environments, facilitating integration with both aerial and ground robotic platforms.

Stay tuned for further developments as this research advances the next generation of intelligent soft-robot systems!