I’m not sure you can say that imitation learning has been under-researched in the past. Imitation learning has been tried before alongside RL. But it did not generalize well until the advent of generative diffusion models.