Robot-based, assist-as-needed (AAN) therapeutic devices are one solution to the high costs and increased demand for rehabilitative therapy. Three decades of work have improved the design of robot-based AAN devices; however, the role that humans play in successful AAN rehabilitation has been less well-studied. In this paper, we briefly describe a framework for understanding human engagement with AAN devices. We then test this framework with a small sample of 10 healthy volunteers. We find evidence that task-based (weight) characteristics and physical states (accuracy) inform peoples’ Judgments of Difficulty (JODs). In contrast, volunteers’ engagement with AAN rehabilitation was tentatively informed by the amount of time spent on the last exercise. Together, these results suggest that JODs and task engagement decisions are distinct – but related – constructs that are present during rehabilitation.