emily x meschke



Back to Top

Learning interruptibility

As people become more reliant on robots to complete basic tasks, establishing methods to enhance collaboration and communication between humans and robots becomes more essential. This branch of my research focuses on developing cognitively informed, computational models of hidden human states (such as availability) that autonomous agents should learn to consider when interacting with people.

For this project, I worked closely with Elizabeth Cha, to model the state of the human and robot as an MDP. Based on the urgency of the robot's request and the availability of the human, the robot learns to determine which non-verbal signal is appropriate for the situation.

I developed a simple, RPG world in Javascript and designed the game tasks to thoroughly test the model. The code for the game is available here.

In the next iteration of the project, I won't assume that the state of the human is known to the robot, modeling the learning problem as a POMDP instead of an MDP.