What is the point of this experiment?

We are creating interfaces that communicate robot learning to human teachers. During this experiment you will physically interact with a robot arm and teach it to do a task. The robot is unsure about different features of the task. We want to test how a robot can intuitively and seamlessly tell you what feature(s) it knows and what feature(s) it is uncertain about. In the long term, we want to make it easier for everyday users to teach intelligent robots.

What am I teaching the robot?

This video shows the task that you are trying to teach to the robot:

The task consists of three features:

Keeping the robot close to yourself
Keeping the robot close to the table
Keeping the end-effector orientation upright

The robot is confused about these features. At different parts of the trajectory it is not sure how to stay close to the table, away from you, and keep the cup upright. It’s your job to identify what the robot is confused about, and then teach that feature.

How will I know what the robot is confused about?

The robot will use three methods for revealing its confusion.

GUI (Computer Monitor)

Here the robot will display three values on the computer monitor. (1) uncertainty about distance from you. (2) uncertainty about distance to the table. (3) uncertainty about the end-effector orientaion.

Local (Haptic Bags)

Here the robot will use the haptic bags wrapped around its arm to reveal information. Notice that there are three bags on the robot arm. Each bag corresponds to a different feature: inflating bag (1) indicates uncertainty about distance from you. Inflating bag (2) indicates uncertainty about distance to the table. Inflating bag (3) indicates uncertainty about the end-effector orientation.

Hint: You may need to move your hand to grab the different bags and check if they are inflated.

Global (Haptic Bags)

Here the robot will use the bags in a different way. Each bag has three columns. Inflating column (1) means uncertainty about distance from you. Inflating column (2) means uncertainty about distance to the table. Inflating column (3) means uncertainty about the end-effector orientation.

Hint: All the bags have the same columns, and reveal the same information.

What do I need to do?

First, you will teach the robot to perform the task without any feedback. We will use this initial demonstration as a baseline.

Then you will teach the robot with each type of feedback:

Once with GUI
Once with Local
Once with Global

When you get feedback from the robot, try to exaggerate the feature the robot is confused about. For example, if the robot signals that it is uncertain about distance to table, focus on moving the robot closer to the table.