Drive the cart

Game

Episode 1

Space launches · hold ← / → (or A / D) to run the motor · let go to coast

State

what an agent would see

Reward per episode

0 episodes
Caught 0
Missed 0
Success (last 20) –
caught missed average of last 20
actions: ← motor left · coast · motor right →
cart: v ← v + (14·a − 0.6·v)·Δt, a ∈ {−1, 0, +1}
coasting adds rolling friction (2 m/s²)
reward: +1 catch · miss by d metres: −min(1, d / 10)
keys are read every 0.05 s, the same rate the Q-learning agent picks actions