Background: Gait adaptability is essential for stroke survivors to achieve efficient and safe community ambulation. However, conventional treadmill rehabilitation is only a repetitive practice of leg movement. This study compared the effects of augmented reality (AR) treadmill-based gait adaptation training with regular treadmill programs for patients with stroke.Methods: Forty patients with stroke (N = 40) were randomly assigned to the gait adaptation training (n = 20, age: 49.85 (SD 8.44) years; onset of stroke: 107.80 (SD 48.31) days) and regular training (n = 20, age: 50.75 (SD 8.05) years, onset of stroke: 111.60 (SD 49.62) days) groups. Both groups completed three sessions of training per week for five weeks (15 sessions). The primary outcomes were the 10-meter walk test, and success rate of obstacle avoidance, while secondary outcomes included the Berg balance scale, component timed-up-and-go, and fall rate in a six-month follow-up period. Assessments were performed before and after the intervention. The paired t-test was applied to compare the differences within groups and independent sample t-test was performed to compare the differences between groups.Results: The 10-meter walk test, success rate of obstacle avoidance, Berg balance scale, component timed-up-and-go all significantly improved in both groups (P < 0.001). The success rate of obstacle avoidance (P = 0.02, 95% CI: -21.07, -1.64), Berg Balance Scale (P = 0.02, 95% CI: -8.03, -0.67), “turning around time” (P = 0.04, 95% CI: 0.08, 2.81), “stand-to-sit” (P = 0.03, 95% CI: 0.16, 2.41) and “total time” (P = 0.048, 95% CI: 0.04, 10.32) improved significantly in gait adaptation training group after intervention, while the 10-meter walk test (P = 0.09, 95% CI: -0.17, 0.01), timed “sit-to-stand” (P = 0.09, 95% CI: -0.14, 2.04) and “linear walking” (P = 0.09, 95% CI: -0.27, 3.25) in gait adaptation training group didn’t show statistical difference compared to the regular training group. Total fall rate during the follow-up period was statistically decreased in the gait adaptation training group (P = 0.045).Conclusions: Both interventions improved mobility outcomes, with AR treadmill-based gait adaptation indicating greater improvement in obstacle avoidance, balance, turning, and stand-to-sit. AR treadmill-based gait adaptation training emerges as an effective and promising intervention for patients with stroke in early rehabilitation.