01 Interactive DQN Simulation
Balance is a moving target.
I rebuilt my original CartPole-v1 work as an interactive browser experience that makes the agent’s decisions visible. Watch it read four changing state values, choose a direction, learn from stored experience, and improve as it completes more episodes.
LIVE ENVIRONMENTReady
CONTROL MODEAI Agent Mode
CartPole simulation
01 / Cart Position0.000
02 / Cart Velocity0.000
03 / Pole Angle0.000°
04 / Pole Velocity0.000
02 Simulation Controls
Run it. Pause it. Push it off balance.
Switch between the trained agent and manual control, then follow the four live state values to see how every left or right movement affects the system.
Manual input is active. Use the arrow keys or press either force control.
Q-VALUE UPDATEImmediate reward + discounted future value
q_update = reward + GAMMA × max(Q(next_state))
03 Learning Telemetry
Six numbers tell the story.
04 Learning Curve
Episode Scores
05 Run Output
Training Log
- Simulation initialized. The agent is ready.