Learning from task related instructions.

Co-constructing an understanding gradually step by step about how to execute the task is crucial for learning and mastering any new skill. We are exploring ways to employ existing robot skills to build new behaviors and perform novel tasks. Task instructions provided by bootstrap coaching language will be created to teach robots new ways of performing novel everyday tasks.

An ITL framework will provide a direct interface of robot skills to be manipulated by human instructions. Such as pickup a milk carton will prompt the robot to explore its picking-up motion skills. Several such skills will be listed for human teachers to directly access via user-friendly tools. Additionally, these task-specific instructions from the human teacher will be saved in a robot belief state as part of an episodic memory and can later be used for performing similar tasks.