Standing starts before walking#

The latest work separates calibration Home from ordinary startup, adds adjustable self-leveling to Body Control, and investigates why a smaller linkage is not automatically easier for the motors to drive.

The October 13 hardware journal records an owner-reported 566 g assembled mass, 5 V supply and MG90 servos. These replace earlier unknown mass and voltage entries for that reported configuration. They do not establish the finished shell’s mass, measured current draw or battery endurance.

Crouch, align the shoulders, then rise#

Ordinary boot previously drove toward saved calibration Home. On the floor, that could extend the legs before the shoulders were aligned. Startup now requests a staged rise from the body controller:

Fold the carrier and crank joints toward a modeled crouch.
Align the shoulders while retaining the crouch.
Raise the body into Stand over the modeled foot contacts.

The later revision staggers the front pair and rear pair by 100 ms, with a total commanded sequence of 2.5 seconds. Calibration Home remains a separate explicit operation. Staggering command starts has not yet demonstrated a measured reduction in peak current.

The firmware journal records Wi-Fi installation and acceptance of the startup revisions, preservation of joint calibration and native tests of phase order, joint mapping, contact geometry and cancellation. These are commanded trajectories: the standard servo interface does not report actual arrival or foot contact. A completed startup command alone cannot prove that the chassis physically rose.

Self-leveling has its own controls#

The P4 now exposes self-leveling settings through the selected robot’s Body Control connection. They are saved on the robot independently of joint and IMU calibration.

Control	Meaning
On / Off	Enables correction in supported standing and moving states; separate from whether correction is currently active
Response strength	Multiplier for the attitude-error response; original default 1×
Damping	Multiplier for the angular-rate response; original default 1×
Roll target	Desired roll offset; positive raises the left side, original default 0°
Pitch target	Desired pitch offset; positive raises the nose, original default 0°

Turning correction off does not stop a gait or disable outputs. Returning to On can resume correction in an eligible state without starting another motion. The controls read their values from the robot; selecting default response values only edits the form until they are saved.

An earlier tilt test found fresh gyro data but inactive balance after a shrug. Selecting Stand activated correction. Further checks found changing reported joint commands while the owner still reported little physical response. The investigation then identified a separate controller issue: the old standing response settled at a fraction of a persistent tilt error rather than continuing to correct it.

Stationary hold now accumulates accepted correction over time. Moving gait and rise retain their separate response. An ideal-actuator test exercises rejection of a sustained disturbance; real linkage fit still determines which corrections can be applied. Increasing a setting does not increase the servo’s available torque.

Installed controls, remaining physical work#

The October 14 journal records 0.8.46-p4-leveling-tuning uploaded, explicitly activated and accepted over Wi-Fi. Live browser checks read and saved the four response values, preserving the previously selected Off state. The recorded session ended in Doze with outputs disabled.

This establishes firmware installation and working configuration controls. Successful loaded self-leveling remains unverified. The IMU measures body attitude; it cannot identify an individual stalled joint or tell a level body standing above the floor from one sagging onto it. An automatic fallen-body get-up routine is also not established by this work.

A calculator for the actual linkage#

The new local leg-geometry calculator keeps the servo shaft locations fixed by default while changing link lengths. It can animate a requested path, compare design snapshots, plot joint travel and modeled load demand, and export JSON, CSV and SVG. It has no connection to robot controls.

The October 14 comparison uses the last-confirmed assembled geometry of 55 / 35 / 38 / 28 / 20 mm for leg, carrier, rod, pickup and crank, versus an active Blender study of 45 / 35 / 30 / 28 / 20 mm. The shorter study reduces reach, but its shorter rod approaches an unfavorable alignment as the body rises. Under the journal’s particular 65 mm shaft-height benchmark, modeled carrier peak torque falls about 3%, while crank demand rises about 16%. That is not a general torque improvement.

The calculator uses ideal point feet and external-force geometry. It omits housing collisions, servo torque-speed limits and whole-body balance. Its plots are design evidence, not a recommendation to change the assembled robot’s calibration or a physical gait qualification.

The Blender work also has a separate motion working file retaining the main 23-chapter motion library and both body configurations. Earlier design scenes remain in the original reference file. The article thumbnail is an October 12 Blender view, not a new photograph or a render of the later geometry study.

Source: the October 13–14 gait research journal, calculator source, production body/balance code and dashboard controls, identified in the source manifest.