ASTRA Simulator Automates Air Traffic Control Training with AI Simpilots.

Training Air Traffic Control Operators (ATCOs) is critical for aviation safety but is often limited by the availability of human "simpilots" who role-play both pilots and other ATCOs in simulated environments. Existing automated training solutions frequently underperform in diverse linguistic contexts, such as with Singaporean-accented aviation speech, where standard speech models show very high word error rates. The ASTRA simulator addresses these challenges by providing an end-to-end AI-driven solution. It automates the simpilot roles through a sophisticated pipeline that transcribes ATCO speech, interprets instructions, and generates realistic pilot and ATCO responses. A key innovation is the use of locally adapted voice models, which significantly reduce the word error rate for specific accents, making the simulation more accurate and effective. Beyond just traffic simulation, ASTRA also integrates an AI-assisted framework for evaluating trainee performance. This system assesses radiotelephony communications based on accuracy, brevity, and completeness, providing standardized and scalable feedback. Built on open-source technologies, ASTRA aims to enhance ATCO training capacity and reduce the workload on human instructors.