Socially assistive robots have been used extensively in designing therapy interventions for children on the autism spectrum. One of the major challenges facing this field is the difficulty in sustaining the child's engagement in a designed activity for longitudinal studies once the novelty of the robot wears off. Another major challenge is the design of intervention strategies that can be adapted to suit the specific needs of every child. This paper presents a novel adaptive framework that addresses both these problems by implementing intervention methods that are tailored to each individual child.