In RL literature this is generally called "curriculum learning".

The curriculum is usually modeled as some form of reward function to steer learning, or sometimes by environment configuration (e.g. learn to walk on a normal surface before a slippery surface).