Legible user input for intent prediction

In assistive teleoperation, the robot provides assistance by predicting the user's intent. Prior work has focused on improving prediction by adapting it to the user's behavior. In this work, we investigate adaptation in the opposite direction: training the user's behavior to the prediction. Results from our user study suggest that users can significantly improve the performance of a simple static predictor after brief exposure to its behavior. In addition, we find this improvement to be more significant when the cognitive load of teleoperation is reduced.