Run the trained model

Compare every controller on the printed arm, then move the model onboard.

Run every controller through one off-board route, then move the learned model and a port of the safety supervisor onto the Pico 2. Publish the hardware table and replay the task ladder, whatever they show.

Written lesson + build labEvidence-based gate

The goal for this build

What you’ll be able to do.

Compare every controller through one off-board route with equal latency.
Port the safety supervisor to the Pico 2 and prove its link timeout on video.
Export the model from JAX to C and verify its outputs, worst-case timing and onboard tracking.
Run the 18-cell matrix and the task ladder, and publish the table whatever it shows.

What to have ready

Module 8 gate passed: every interval inside its training range, or the model retrained
The Module 5 policy and adaptation module, at most 10k float32 parameters, with their normalization
The stock loop, PID, gravity feedforward and the EKF from Modules 2 and 3, frozen for the benchmark
The second Pico 2 on the harness’s Pico 2 tap

Download the course starter ↓

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, the task and the gate command for this module, and to download the starter kit.

Create a free account Sign in

03 / Prove it

Keep the evidence.
Then run the gate.

What you ship

Logs of every run, with commit, config and seed
The hardware bench.py table for every controller and cell, published whatever it shows
The C export with its parity and worst-case timing records, and the Round 2 failsafe video
The scanned drawings overlaid per controller
A video of every task on the ladder

Module 09 gateEvidence required

The hardware table is complete for every controller and published, whatever it shows.
Onboard outputs match JAX within 1e−4 on recorded inputs.
Onboard RMS error stays within 10% of the off-board result, beyond run-to-run noise.
Inference finishes within 2 ms worst case on the Pico 2.
The Round 2 failsafe test passed on video: the Module 7 failsafe test repeated on the Pico 2, plus the 0.2 s link timeout with the USB cable pulled.
The learned model never causes a safety latch, and every latch has a written cause.
The dock task passes 10 of 10 cycles within the Task ladder’s current limit: no safety latch, and the shoulder’s RMS current per cycle under its rated-torque current (about 0.9 A).
A tuned classical controller may win on hardware, and that is a valid result.

With a free account: the gate command and how to record your evidence.

The story behind the build

Episode companions.

EP 10The model meets the printed arm EP 11My arm draws, weighs, throws and docks a drone

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.