Part 01 / Software & machine learning
03 / 09The analytical fix
Estimate the physics before learning it.
Augment an EKF with mass and a disturbance force, then feed those estimates back into the classical controller. This is the adaptive baseline the learned policy must earn its way past.
The goal for this build
What you’ll be able to do.
- Implement EKF prediction and measurement updates.
- Explain why some parameter combinations are unobservable.
- Use estimated mass and disturbance in bounded feedforward.
- Judge estimates through tracking, residuals and uncertainty.
What to have ready
- Module 2 benchmark and PID
- Covariance and Gaussian uncertainty basics
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
- Augmented EKF with documented uncertainty assumptions
- Observability experiment and estimate plots
- PID versus EKF-adaptive benchmark comparison
- Module 03 evidence
Module 03 gateEvidence required
- From a prior at least 10% off, every mass estimate after the 2 s identification window is within 5%, with 2σ within 5%.
- After the window, EKF tracking in the +20% mass condition stays at most 1.5× nominal RMS or nominal plus 5 mm, whichever is larger, with no EKF crash or abort in the nominal or +20% mass runs.
- The thrust calibration, excitation, estimation uncertainty and a PID comparison with adaptation off are documented.
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.