Stroke survivors must overcome motor deficits that greatly impede their balancing ability, thus stunting their independence and overall quality of life. Robot-aided rehabilitation is a new approach to balance therapy presenting notable advantages in efficiency, precision, and consistency. Prior studies have demonstrated the success of visual feedback, force plates, surface perturbations, and compliant surfaces in improving balance control for people with neuromuscular disorders. However, this study is the first to investigate the effect of incorporating each aspect into a stroke balance training program. The side-specific robotic platforms used could generate perturbations while also simulating compliant surfaces. During the 6-week study, 2 subjects each completed 9, 2-hour long training sessions, along with a clinical pre- and post-assessment. Subjects utilized visual feedback of center of pressure and weight distribution to strive for successful balance, and the platforms perturbed if balance was maintained for a sufficient time period. To increase training difficulty, platform stiffness decreased with increased performance. Improvements in functional postural balance for both subjects were demonstrated by the Berg Balance Scale, Mini-BESTest, Timed 10-Meter Walk Test, and 5 Times Sit-to-Stand Test. Decreases in Time to Perturb and Time to Stabilize were suggestive of improved dynamic postural balance. Decreased platform stiffness indicated sustained improvements in increasingly challenging environments, and a 3-month follow up revealed retained functional balance improvements. These results demonstrate the effectiveness of patient-adaptive perturbation-based robotic training on compliant surfaces in improving postural balance for chronic stroke patients.