["European Journal of Education, Volume 61, Issue 3, September 2026. ", "\nABSTRACT\nPersonalised training games have proven to be highly beneficial for children with autism spectrum disorder (ASD) due to the adaptive nature of the interventions in custom training games using reinforcement learning (RL). However, their wider applicability and implementation are hindered by issues such as low scalability, inability to adapt to real‐time changes, small training datasets and reliance on costly specialised equipment. A new framework to address these limitations is proposed using Proximal Policy Optimisation (PPO), a recently developed RL approach that dynamically updates game interventions based on real‐time measurements of participants, such as behavioural measures and physiological signals. This approach shows significant improvements in performance, with an engagement prediction accuracy rate of 98% for E4AccData and Performance Data. Additionally, the involvement and performance of participants during intervention sessions show substantial increases, indicating that adaptive training games can effectively promote better learning outcomes for children with ASD. This study contributes to the field by proposing an evidence‐based intervention that can transform the educational experiences of children with ASD and be practically applied in classroom settings.\n"]