Manipulators Adaptability in Uncertain Scenarios for Dynamic Tasks

Robotic systems are becoming increasingly complex and are now expected to operate beyond structured scenarios, where challenging tasks and uncertainties are the norm rather than the exception. Such conditions demand control strategies capable of adapting to unknown system dynamics and partially defined surroundings. This thesis investigates the integration of adaptive control techniques and system compliance to improve the adaptability, robustness, and safety of robotic manipulators, while simultaneously enhancing performance in dynamic tasks. Particular emphasis is placed on the role of feedforward actions and elasticity. The former enables the use of an internal model to accurately follow the desired task. The latter, considered both as an intrinsic mechanical property of the robot and as a feature embedded within the control architecture, allows the system to exploit its natural dynamics. Within this context, the throwing task is selected as a representative case study, as it constitutes a highly dynamic manipulation skill that has recently attracted increasing interest in logistics, recycling, and domestic service applications. Throwing provides an effective benchmark for analyzing the interplay between adaptive learning, compliant interaction, and dynamic trajectory generation. Starting from well-defined industrial scenarios, this research extends its findings to semi-structured and domestic environments, where uncertainties in the robot, perception, and object properties challenge conventional control assumptions. The proposed methodologies contribute to the design of adaptive and compliant robotic systems capable of adjusting their behavior in response to environmental needs, bridging the gap between industrial precision and domestic flexibility.