Watch how an AI learns to explore and find the best reward.
In this video, we enter a simple grid world where an agent navigates through tunnels in search of treasure. There are two options: a small reward close to the starting point, and a larger reward that requires a longer and more uncertain path.
At first, the agent often settles for the easy win. This leads to a suboptimal solution, where it repeatedly chooses the smaller treasure instead of discovering the better one. The key challenge is exploration.
Through intuitive animations, we show how the agent balances trying new paths with using what it already knows. At the same time, the agent’s internal memory is visualized and evolves step by step. You will see how its learned values change as it gains experience, and how this memory eventually guides it toward the better choice.
đź“ş This video is part of the series:
Reinforcement Learning Explained – Animated Examples
Theory is excluded in this version