Human-Inspired Robot Control for Proactive Physical Human-Robot Interaction

Human-human collaboration is characterized by a back-and-forth, where the action of one agent elicits the response of the other. Interaction is inherently multi-modal and includes both high-level modalities such as language and low-level ones such as force exchanges. For a robot to be functional in a human environment, it is essential that the behavior of the robot feels comfortable to the human. That is, robots must perceive clues from all modalities, generate legible action, take initiative, and follow human initiative when necessary. This thesis explores how the robot can be controlled to act proactively in physical human-robot interaction tasks. Our study focused on a specific aspect of physical interaction: negotiation on the direction in collaborative object relocation. We carried out two experiments to observe how participants use the haptic sense alone to agree on which way to move. Both qualitative and quantitative analyses revealed consistent patterns in the haptic and kinematic signals that indicate intent. These patterns were used to develop a framework that can predict human intent in real time. Building on strategies derived from human interaction, we propose a high-level robot controller. The controller utilizes the human intent recognizer and sends robot actuation commands depending on the state of interaction. It functions at a symbolic level, demonstrating the feasibility of using the proposed intent recognizer in real-time control. The human-robot study reveals that the robot accurately identifies and reacts to human actions, clearly conveys its intentions, and resolves conflicts. We claim that this controller can incorporate verbal and gestural communication. We believe that this work will be a stepping stone toward achieving a domestic robot capable of physically collaborating across a wide range of modalities.