Background: Robot-assisted arm training (RAT) is an innovative exercise-based therapy that provides highly intensive, adaptive, and task-specific training, yet its effects for stroke individuals with unilateral spatial neglect remain to be explored. The study was aimed to investigate the effects of RAT on unilateral spatial neglect, arm motor function, activities of daily living, and social participation after stroke. Methods: In a pilot randomized controlled trial, individuals with unilateral spatial neglect after right hemisphere stroke were equally allocated to intervention group and control group, 45-min training daily, 5 days/week, for 4 weeks. Outcome measures included the Behavioral Inattention Test-conventional section (BIT-C), Catherine Bergego Scale (CBS), Fugl-Meyer Assessment for Upper Extremity (FMA-UE), Modified Barthel Index (MBI), and World Health Organization Disability Assessment Schedule Version 2.0 (WHODAS 2.0). Results: From November 2018 to February 2021, 20 stroke patients (mean age 47.40 +/- 8.47) were enrolled in the study. Robot-assisted arm training was feasible and safe for individuals with unilateral spatial neglect. Both groups had significant improvements in all outcome measures. Participants assigned to RAT therapy had significantly greater improvements in BIT-C (difference, 7.70; 95% CI, 0.55-14.85, P = 0.04), FMA-UE (difference, 5.10; 95% CI, 1.52-8.68, P = 0.01), and WHODAS 2.0 (difference, -7.30; 95% CI, -12.50 to -2.10, P = 0.01). However, the change scores on CBS and MBI demonstrated no significance between the groups. Conclusion: Our findings provide preliminary support for introducing robot-assisted arm training to remediate unilateral spatial neglect after stroke. The training program focusing on neglect of contralateral space and affected upper extremity may be effective in neglect symptoms, motor function recovery, and social participation, while not generalizing into improvements in activities of daily living.