Part 01 / Software & machine learning
05 / 09Adaptation without knowing the physics
Train with privileged information. Fly without it.
Implement a two-stage RMA-style policy, then compare it with a policy using the frozen predictive encoder. Keep the compute, observation and deployment budgets visible.
The goal for this build
What you’ll be able to do.
- Train a privileged policy with domain randomization.
- Distill its latent from causal on-policy history.
- Compare RMA and predictive-latent Variant B fairly.
- Measure recovery, parameter count and actual compute cost.
What to have ready
- Module 4 encoder and data discipline
- Module 3 adaptive baseline
- A working vectorizable simulator
Free account
Read the lesson, run the lab, pass the gate.
The whole course is free. Create an account to unlock the lesson, the build lab and the gate command for this module, and to download the starter kit.
03 / Prove it
Keep the evidence.
Then run the gate.
What you ship
- Privileged teacher and causal adaptation module
- Variant B policy using the frozen predictive encoder
- Held-out benchmark including all classical baselines
- Compute report, deployment parameter count and Module 05 evidence
Module 05 gateEvidence required
- In every held-out condition and trajectory the learned policy has no crash or abort, and its RMS is at most 1.10× the EKF’s plus 5 mm.
- Summed over the adaptation conditions, every perturbation except sensor noise and latency, learned RMS does not exceed the EKF’s.
- Every learned-policy mass-step run recovers within 1 s without a crash or abort under the fixed recovery definition.
- RMA and Variant B are reported side by side.
- Deployed networks contain at most 10,000 float32 parameters, and recorded CPU training time is below two hours. This is a pass criterion, not a promised runtime.
With a free account: the gate command and how to record your evidence.
The story behind the build
Episode companions.
Written companions are available now. Recordings and customer access are forthcoming.
Made it through the lesson?
Save your reading progress here. Record your gate result in your project.