Learning human-robot collaboration with POMDP

Modeling and planing a collaborative task for a human and a robot in an unstructured environment is a challenging problem. Most of earlier works simplify the task by assuming exact knowledge of the environment and the human intention, little work has been done on how to plan a collaborative task in the presence of uncertainty. However, there still exists a lot of challenges because of limited perception abilities. To reason explicitly about uncertainty during collaboration between a human and a robot in a household task, we here present a decision making approach for human robot interactivity based on a Partially Observable Markov Decision Process (POMDP). The approach is validated in simulation in a collaborative scenario of assisting a person make a cup of coffee.