From Spain to Italy to advance learning-based control of legged robots

From early to late June 2026, Matilde Puricelli, researcher at the Centre for Automation and Robotics (CAR) of the Spanish National Research Council (CSIC), carried out a secondment at the Department of Information Engineering of the University of Pisa, within the NeutraWeed Project.

Her stay focused on the development and comparison of control strategies for legged robot locomotion, with particular emphasis on gait trajectory tracking. The research explored traditional and learning-based approaches, focusing on Iterative Learning Control (ILC), a technique that enables robots to learn from tracking errors and progressively adapt their movements across repeated iterations.

During the secondment, Matilde first deepened her knowledge of the theoretical foundations of ILC and attended a PhD course on Optimal Control, broadening her understanding of modern control techniques. She then applied this knowledge to the simulated locomotion of a bipedal robot, comparing different control strategies and evaluating their ability to track reference trajectories while adapting to disturbances caused by ground contact.

The research also moved from simulation towards real-world implementation. An experiment was carried out using an available legged robotic platform equipped with Series Elastic Actuators (SEA), testing the integration of an ILC-based controller and identifying the challenges involved in transferring the approach from simulation to physical hardware.

The secondment also provided valuable opportunities for scientific exchange. Matilde participated in the weekly research group meetings and attended seminars on topics including control-learning fusion for compliant robots and human-robot interaction and assistive technologies. These activities fostered discussions with researchers working on different robotic systems and control approaches, while creating opportunities for new professional connections and future collaborations.

Preliminary results indicate that the ILC framework can effectively reduce trajectory tracking errors by learning from disturbances introduced by ground contact. Experimental results also showed that the controller can be successfully integrated with the available robotic platform, providing a promising basis for further validation under realistic operating conditions.

Matilde’s secondment strengthened collaboration between CSIC and the University of Pisa, contributing to the development of adaptive and learning-based approaches for legged robot locomotion and supporting the exchange of expertise across the NeutraWeed research network.

Stay tuned for more updates from the NeutraWeed international research community!