Goals vs. Actions as User-Facing Representations for Robot Programming

Abstract

Robot application development (RAD) tools provide novice developers with the ability to specify what tasks a robot needs to perform and how it should perform these tasks. Existing RAD tools often focus on how end-user developer intent should be captured and stored as a computational artifact, including but not limited to mixed-reality interfaces, natural language programming, and visual programming environments. However, these tools fail to incorporate robot intelligence in the development pipeline, making it cumbersome for end-user developers to specify a detailed sequence of steps for the robot to complete a task, as well as overconstraining the robot. We thereby suggest that robot intelligence needs to be incorporated into RAD tools. Our proposed work will focus on balancing the amount and nature of domain knowledge that the end-user developer must provide to the robot with the robot’s ability to plan and act autonomously, as well as how the end-user’s domain knowledge should be represented as a computational artifact. In particular, we will be comparing two different approaches – goals versus actions – as a user-facing representation for collecting developer task specifications and intent while leveraging the robot’s ability to plan and act autonomously through robot intelligence by extending the Polaris interface, developed by Porfirio, Roberts, and Hiatt (2024), through a user study.