Older adults generally engage in less physical activities than other age groups, which can increase the risk of developing diseases associated with aging. It is increasingly difficult for caregivers to support older adults due to the global decline in the old-age support ratio. Socially assistive robots have shown success in supporting older adults and extending the capability of healthcare workers. This thesis aims to develop:​ 1) an affect elicitation and detection methodology using a social robot to directly elicit user affect that occurred during human-robot interactions (HRIs) to train affect detection models;​ 2) an autonomous socially assistive robot to facilitate upper body exercises. Exploratory experiments with both younger and older adults were conducted to validate the robot’s ability to elicit user affect and facilitate exercises. The results showed that the robot successfully elicited and detected user affect as well as facilitated exercising, and that the participants also reported high acceptance of the robot.