Andrea Protopapa

The objective of this research is to advance robot learning for compliant and adaptive behaviors in dynamic and unpredictable environments. The investigation focuses on the development of effective and efficient learning strategies for complex robotic systems, addressing challenges associated with real-world complexity, task modeling, and domain transfer. Key contributions include the development of domain randomization techniques for robust deployment of real-world policies in soft robotics and the introduction of video-based inverse reinforcement learning to simplify complex task definitions in robotic manipulation. By leveraging graph-based abstractions, the research points to enable robots to generalize across diverse scenarios and tasks. This work aims to develop robust, adaptive robotic systems capable of operating efficiently in real-world applications.