The research question addresses how to prevent human errors in physical assembly tasks, particularly when users have flexibility in choosing the construction order.

TanCRAFT consists of a four-stage real-time system that monitors the scene, reasons about the current assembly state, predicts future piece movements, and intervenes if the predicted placement is incorrect.

The authors report the effectiveness of TanCRAFT through offline benchmarks and a user study in a tangram assembly task, showing it reduces mistakes and improves efficiency compared to unguided and reactive approaches.