Is Retard Learning Better Than Supervised Learning?

The agent learns about colors and shapes of objects by trying to pick the object that has a certain color and shape given a command like, find the red triangle. And you just get a strong enough signal from the supervised learning, i think, early before the reinforcement learning starts to take over. So this is part of my intuition for why this might work better for a reinforcement learning agent than for a supervised learning set up. You get stronger signals from the supervised explanations early in learning, before you start really solving the task.