← AI-aktiviteter Svenska

Q-learning simulation

Train an agent to navigate a grid with reinforcement learning. Watch the Q-values update in real time.

Agent Goal Coin Obstacle
Episode: 0, Step: 0
Speed
Parameters, which can be changed any time, even during training
α learning rate
0.50
γ discount factor
0.95
ε exploration rate
0.90
How the agent chooses an action right now:
Exploration 90%Exploitation 10%
Step log · click a row for an explanation
Learning curve · number of steps to reach the goal per episode
Start training and the curve will be drawn: the number of steps per episode drops as the agent learns a shorter path.
Learn more · click for an explanation

Q-learning simulation

An agent moves through a grid and has to learn to collect all the coins and reach the goal, just by trial and error and receiving rewards and penalties.

1
Train the agent. Press Start training, and the agent starts trying its way through the grid.
2
Follow the Q-values. The color and numbers on each cell show what the agent has learned so far about each action.
3
Adjust the parameters. Change α, γ and ε and see how it affects how fast and stably the agent learns.