Cross-Task Knowledge Transfer for Visually-Grounded Navigation