First stand & first steps

First motion is an experiment, not a demonstration

Download test-card.csv and joint-map.csv. Complete the test card before applying motor power: configuration, purpose, allowed command envelope, duration, observations, stop conditions and recovery. The exact values must come from your verified hardware limits and commissioning results. The course does not establish a universal safe gain, walking speed or current for every BDX-R build.

This Academy progression assumes the previous module passed. If a motor identity, sign, zero, limit, sensor mounting or stop behavior remains uncertain, return to that check. A controller that walked in the supplied simulation has not thereby been validated on your mass distribution, mechanics or harness.

Understand what each stop action actually does

A zero velocity request can ask a walking policy to stand; it does not necessarily stop joint motion. A software pause can leave a last target active. Disabling torque can let the robot collapse. Electrical isolation removes power through the designed path and can have its own mechanical consequences. Verify these behaviors with support before relying on them.

Use a fixture and restraint arranged for the assembled robot, a clear test area and a second person when the recovery procedure requires one. Keep people outside moving pinch points and the possible fall path. Operate the defined stop from outside that space; do not plan to catch a powered robot by hand.

Support must be measured honestly

Begin with the body supported and feet clear as appropriate to the completed commissioning procedure. Progress to feet contacting the floor with the support still able to prevent a fall. Check cable clearance and contact geometry before asking the legs to bear more load.

A taut tether or a fixture pushing the torso can make standing look successful. Record whether it carries load. Independent balance requires the feet and controller to maintain posture under the intended test conditions while the protective restraint remains inactive. Never remove protection simply to obtain a better video.

Use the stages in order: passed joint and sensor commissioning, supported whole-body checks, a documented standing result under the intended support conditions, and only then a small walking request. Each stage answers a different question. A successful simulated gait or a completed cosmetic assembly cannot substitute for the preceding physical checks.

Separate diagnosis from tuning

Log timestamped state, joint positions/velocities, commands, orientation, available current/temperature data and communication faults. Synchronize a camera view with a visible event or timestamp. Compare expected and observed behavior rather than relying on how smooth the clip feels.

If the robot leans, oscillates or steps unexpectedly, first check configuration, calibration, sensor frame, contact and support interference. Stop the trial before investigating. Gain changes can hide a sign, delay or mechanical problem without resolving it. Later system-identification and simulation-training modules provide a controlled route for changing the model and policy.

If two settings changed before a failure, the result does not tell you which change caused it. Keep the failed log, recover using the established procedure, and restore the last accepted configuration. Then test one hypothesis at a time against that baseline. Changing both settings again makes the cause harder to identify.