Computational thinking (CT) is an important competency for students in the 21st century, but current approaches to promoting students’ CT development have limitations. This study aims to propose a humanoid robot-assisted collaborative programming (HR-ACP) approach and examine its effects on students’ CT, learning motivation, and interaction. A quasi-experimental study was conducted in a middle school programming course. 45 students were randomly divided into the HR-ACP group (N = 22) and the conventional block-assisted collaborative programming (CB-ACP) group (N = 23). The results show that the HR-ACP approach significantly improved students’ CT concepts and learning motivation than the CB-ACP approach. It is worth noting that the improvement in CT practices was not significant. In addition, students in the HR-ACP group demonstrated deeper human-human and human-machine interactions compared to those in the CB-ACP group. Based on the HR-ACP approach and findings in this study, some possible reasons and specific practical recommendations are further discussed to better empower students to succeed in the 21st century.