Thumb movement is critical for human hand function and often impaired by stroke. Despite the importance of the thumb, there are few assessments for evaluating the sensory ability of the thumb. The primary objectives of this thesis were to 1) develop a novel robotic assessment of thumb proprioception; 2) validate the assessment with neurotypical participants, focusing on how assessment parameters affect the measurement; and 3) validate the assessment with individuals who have had a stroke across multiple testing sessions. The thumb proprioception assessment we developed centers on a simple video game, called the Circle Jump game, which we implemented with the FINGER rehabilitation robot. In Circle Jump, the robot moves the visually-hidden thumb in a circle, and participants are prompted to press a button when the thumb aligns with a target location presented on a screen, using only proprioceptive feedback to estimate the actual thumb position in comparison to the target location. Two experiments were conducted to assess thumb proprioception using this game. The first experiment involved 26 neurotypical participants who engaged in the Circle Jump task six times in a single session, experiencing variations in speed, direction, workspace size, and the employed finger (thumb or index). The second experiment included 17 stroke survivors who engaged in the task seven times over a 2-month training program to measure potential improvements over time. Within this 2-month period, they also trained finger and thumb proprioception using other robotic games 3 times per week for 3 weeks. For the unimpaired participants, workspace size had little effect on thumb proprioceptive accuracy. Playing Circle Jump at a higher speed or employing the index finger to play modestly decreased error. In contrast, the history of exposure to a direction of rotation had a major effect on proprioceptive accuracy. Specifically, proprioceptive error exhibited a large and transient increase when the rotation direction was reversed after prolonged training in the original direction, mirroring patterns of motor adaptation that have been observed for reaching movements under external force fields. A new proprioception learning model is presented to account for this novel form of sensory adaptation. In the case of stroke survivors, mean thumb proprioception errors were approximately double those for the neurotypical population. Errors remained stable across the seven assessment sessions, showing only a small, nonsignificant improvement over this time. We conclude that the Circle Jump assessment is a valuable new tool for quantifying thumb proprioception. With it, we uncovered a novel form of sensory adaptation, quantified the effect of stroke on thumb proprioception, and found that a 3-week course of robotic training does not significantly improve thumb proprioception.