Ball Thrust 3D: from Rapier to a trainable DQN
Real physics, 21 observations, a DQN that learns and verifiable inference after a local checkpoint reload.
Read the tutorial →
The Ignition blog
Learn by building
Reproducible experiments, code, and agents in action.
Robot Arm 3D: reaching an XYZ target with DQN
An ideal 3-joint arm, 325 configurations and 6 actions: follow how a dense reward brings the tip closer to the target, with real measurements and their limits.
Read the tutorial →
MountainCar: learning to build momentum
Understand why backing up can help an agent reach the summit.
Read the tutorial →
GridWorld: values and actions
Visualising Q-values on a grid to understand what the agent learned.
Read the tutorial →
CartPole 3D: reading the observations
From raw state vector to agent decision in the 3D scene.
Read the tutorial →
Drone navigation: stable control
How the drone demo keeps flight stable while reaching its target.
Read the tutorial →
Maze: sensors and rewards
Dense rewards and sensors guiding the agent out of the maze.
Read the tutorial →
Maze 3D: keys and doors
Exploration, keys and doors in a 3D maze.
Read the tutorial →
Circuit racing: training a driver
Teaching an agent to follow a track with dense rewards.
Read the tutorial →
Train your first agent with Ignition on CartPole
Run your first DQN training in the browser and watch its decisions on CartPole. Executable code and real screenshots included.
Read the tutorial →