TRANSIC is a data-driven approach for sim-to-real policy transfer using human-in-the-loop correction. The method enables robots to learn from human assistance to close simulation-to-reality gaps. It involves training a base policy in simulation, then deploying it on real robots with human intervention for correction. Human corrections are used to train a residual policy, which is then integrated with the base policy to achieve successful real-world execution. TRANSIC effectively addresses various sim-to-real gaps, including perception, embodiment, controller, and dynamics mismatches. It scales with human effort and demonstrates robustness, safety, and generalization to new tasks. The method outperforms traditional sim-to-real approaches and requires less real-world data. It is effective for complex, contact-rich manipulation tasks such as furniture assembly. Videos and code are available at transic-robot.github.io.TRANSIC is a data-driven approach for sim-to-real policy transfer using human-in-the-loop correction. The method enables robots to learn from human assistance to close simulation-to-reality gaps. It involves training a base policy in simulation, then deploying it on real robots with human intervention for correction. Human corrections are used to train a residual policy, which is then integrated with the base policy to achieve successful real-world execution. TRANSIC effectively addresses various sim-to-real gaps, including perception, embodiment, controller, and dynamics mismatches. It scales with human effort and demonstrates robustness, safety, and generalization to new tasks. The method outperforms traditional sim-to-real approaches and requires less real-world data. It is effective for complex, contact-rich manipulation tasks such as furniture assembly. Videos and code are available at transic-robot.github.io.