Handwriting is a critical skill for children during early education, underpinning essential activities such as examinations, note-taking, and self-expression. Among the various factors that influence handwriting development, educational practices commonly emphasise physical behaviour, particularly body posture, although currently no conclusive scientific evidence supports the existence of such a connection. This thesis thus firstly quantitatively examines the relationship between body posture quality and handwriting quality in children, addressing the prevalent assumption that improved posture positively affects handwriting performance. However, regulating the physical behaviour of children, such as body posture, remains challenging. Existing posture regulation systems typically suffer from rapid decay in efficacy over time and are neither suitable for children nor applicable to handwriting contexts. To address these limitations, the overarching objective of this thesis is to explore effective, minimally intrusive and child-centred approaches to regulate the physical behaviour of children, with the long-term goal of supporting habitual physical behaviour change. Building upon a social robot-assisted handwriting training system with tablets - iReCheck, two complementary posture regulation approaches are proposed and evaluated in this thesis: one leveraging visual stimuli on the tablet and one leveraging the social influence of the robot. The former, named WriteUpRight, is an interactive system that employs subtle, slowly deforming visual stimuli on a tablet screen to prompt posture adjustments. Particularly, the approach seeks to induce self-reflection and self-correction by unobtrusively amplifying the postural error of children. It further incorporates personalised and adaptive intervention strategies through reinforcement learning, enabling tailored support aligned with individual children's behavioural patterns and progress. The latter approach, Child-Robot Relational Norm Intervention (CRNI), co-designed with children, utilises social norms by leveraging children's reluctance to disturb others. Unlike traditional posture correction methods reliant on direct reminders, CRNI fosters children's self-monitoring and correction by having the robot express mild distress when the child falls into a poor posture. Together, these contributions illustrate the effectiveness and potential of combining unobtrusive visual interventions, adaptive reinforcement learning techniques, and socially persuasive robotics in tablet-based educational environments to support children's posture self-regulation and handwriting skill training.