In human-robot teams, humans issue control signals in contexts ranging from teleoperation to instruction to shared autonomy. These signals may be used in a variety of ways, including input to inference engines, cues for autonomous intervention, or feedback mechanisms. Generating these control signals requires the physical actuation of an interface, whether via overt body movements or electrical signals from muscles or the brain. However, deviations between the human’s true intended signal and the signal measured by the robotics system can propagate through the control system, undermining performance and imposing an artificial upper limit on team synergy and success.

In this work, we introduce a framework to model, novel algorithms to enable, and extensive human subject studies to evaluate interface-aware robotic intelligence and assistance. By accounting for the unique characteristics of the human input through various interfaces, this body of work demonstrates both the need for and the utility of interface-awareness in enhancing robot teleoperation and autonomy.