This paper will describe the NJIT-RAVR system, which combines adaptive robotics with complex VR simulations for the rehabilitation of upper extremity impairments and function in children with CP. The feasibility of this system is examined in the context of two pilot studies. The NJIT-RAVR system consists of the Haptic Master, a 6 degrees of freedom, admittance controlled robot and a suite of rehabilitation simulations that we have developed. The system provides adaptive algorithms for the Haptic Master, allowing impaired users to interact with rich virtual environments. All subjects trained with the NJIT-RAVR System for one hour, 3 days a week for three weeks. The subjects played a combination of four or five simulations depending on their therapeutic goals, tolerances and preferences. Subjects differed in the level of activity performed outside of NJIT-RAVR system training. Each group of subjects performed a battery of clinical testing and kinematic measurements of reaching collected by the NJIT-RAVR system. Both groups improved in robotically collected kinematic measures and the Melbourne Assessment of Unilateral Upper Limb Function.