Curriculum Computer Science and Innovation for Societal Challenges, XXXVII series  Grant sponsor Università Degli Studi di Padova Supervisor s Lamberto Ballan Co-Supervisor s Gianluca Campana Contact enrico.cancelli@studenti.unipd.it

Project description
Autonomous agents that move and interact with the physical world, like robots and self-driving cars, have to face the important challenge of interacting with human beings in highly dynamic and unknown settings, without causing them any harm. Learning to understand our surroundings while being mindful of others’ intentions and actions is a fundamental requirement to achieve our goal effectively and it’s a fundamental skill for intelligent agents integrated with our world. Current Embodied AI research leverages Reinforcement Learning and Deep Learning to carry out basic navigation tasks in complex, but fundamentally static, indoor environments. My goal is to add dynamic elements in Embodied AI scenes and simulators and to design agents that integrate human behavior prediction skills with navigation and environmental understanding skills. This might lead to real-world deployment of the agent or to the creation of more complex tasks.