Robotic driven treatment plans targeting isolated joints of the upper limb have improved the sensorimotor condition of patients with stroke. Similar intensive efforts to allay lower limb gait impairment have not been so successful. In patients with stroke, targeted robot assisted training of the ankle joint, in a seated position, has demonstrated significant alterations in ankle stiffness and foot-ankle orientation at foot strike which may provide a new treatment option for gait impairment. To determine if isolated robot-assisted training of the ankle joint improves chronic hemiparetic gait in patients with stroke who are categorized according to baseline gait impairment. Patients with chronic stroke (>6mo) and hemiparetic gait (N = 29) received 18 sessions of isolated robot-assisted motor training of the ankle (3×/week for 6 weeks). All participants had stable clinical baseline scores across three admission measures, and no participant was receiving simultaneous outpatient rehabilitation. Baseline gait speed determined three impairment groups: high, >0.8 m/s; medium, 0.4-0.8 m/s; low, <0.4 m/s. Outcome measures included the Berg Balance Scale, the 6 Minute Walk Test, and the 10 Meter Walk Test, and were recorded upon admission, discharge, and 3 months following intervention. Three distinct and significant between-group patterns of recovery emerged for gait speed. The within-group analysis showed that the medium and high group exhibited significant improvements in gait speed and endurance upon discharge, that were maintained at 3-months. Gait speed improvements were clinically significant (>0.16 m/s) for the high function group across all gait speed and endurance measures at discharge and at 3 months. The moderate group also exhibited clinically significant improvements at follow-up on the 10 Meter Walk Test, fast pace (0.16 m/sec), and approached clinical significance for the 10 Meter Walk comfortable pace (0.12 m/sec). The low group had small but significant improvements, at discharge on two of the three gait measures, and these improvements were maintained at 3 months. For balance measures, the low and moderate impairment groups had significant improvements at discharge that were robust on follow-up measure. The high function group demonstrated no significant change in balance. Joint-specific robotic training of the paretic ankle provided the most benefit to individuals with moderate or mild gait speed impairments after stroke. Baseline gait speed function (low, moderate, high) was associated with three distinct recovery profiles. This suggests that severity-specific intervention may be critical to improving efficiency of stroke recovery.