Human-Robot Interactive Planning Experiment
Before the training session, all participants watched a short tutorial on the human-robot interactive planning interface.
Participants of Group A watched the Human-Robot Interactive Planning Using Cross-Training tutorial.
Participants of Group B watched the Human-Robot Interactive Planning Using Reinforcement Learning with Human Reward Assignment tutorial.
After viewing the tutorial, participants of Group A cross-trained with the robot in a virtual environment. In the following videos, we show an example of the Human-Robot Interactive Planning Using Cross-Training session.
People of Group B trained with the robot by assigning rewards. In the following video, we show an example of the Human-Robot Interactive Planning Using Reinforcement Learning with Human Reward Assignment session.
After the training session, we elicit human-robot mental model similarity by asking all participants to run the Human-Robot Mental Model Elicitation Tool. In the following video, we assume that the human thinks that the robot should drill a screw as soon as it is placed, in a sequence from right to left.
Participants then executed a sequence of the place-and-drill task with Abbie in the actual environment. We then compare teaming quality between the two groups using teamwork fluency metrics. We use the subset of participants that had as preference the efficient strategy to “have Abbie drill a screw as soon as it is placed”.
Sample video of a human subject of Group A.
Sample video 1 of a human subject of Group B. Although Abbie learned the human’s preference, the participant is not sure of what the robot will do next and is not confident in doing a placing action while the robot is moving.
Sample video 2 of a human subject of Group B. Abbie did not learn correctly the human’s preference, and this affected teamwork fluency metrics such as concurrent motion and human idle time.