THE MECHANISM

Latent dynamics and imagined rollouts

03Compress observations into state
→
04Predict through actions
→
05Learn in imagination, verify in reality
FAILURE → REPAIR
Common trap

“A good-looking video prediction is automatically a good control model.”

Better model

A control model must preserve action-conditioned, reward-relevant details—even if its decoded pixels are not photorealistic.

YOUR NEXT GATE

Dream a tiny world

  1. 1Use Pendulum-v1 or the referenced Dreamer tutorial.
  2. 2Collect observation, action, reward, and termination tuples.
  3. 3Fit an encoder/dynamics/reward model.