Open-Ended Learning for Interactive Robots (OLIVER)

Funding type: Euregio

Duration: 01/09/2019 – 31/08/2022

For the vision of multi-purpose helper robots to come true, a key open problem is robot skill and knowledge acquisition.  We would like to be able to teach robots to perform a great variety of tasks, including tasks not directly foreseen by its designers.  Thus, the space of potentially-important aspects of perception and action is by necessity extremely large, since any aspect may become important at some point in time.  Conventional machine learning methods cannot be directly applied in such unconstrained circumstances. Thus, a central problem the robot needs to solve is to understand which aspects of a demonstrated action are crucially important and which can be neglected.  Such understanding allows a robot to perform a task robustly even if the scenario changes, to adapt and alter its strategy if necessary, to judge its success, to recognize its own limitations, and to ask for help in specific ways. Moreover, understanding the essence of a demonstrated task allows the robot to infer the intent and progress of humans performing a known task, enabling it to offer help, resulting in natural human-robot cooperative behavior. These capabilities are indispensable for robots deployed in unstructured environments, but constitute largely unsolved scientific problems, which we aim to address in the context of this proposal.

Photography: credits NOI Techpark/Daniele Fiorentino