Better motion tools for the twelve-servo body#

V2 now has revised locomotion paths, a faster seated Wave and more informative self-leveling controls. The October 15 journal records Wi-Fi installation and boot verification through 0.8.52-p4-wave. That establishes deployed firmware; the new gait performance and expressive motion still need loaded physical tests.

The revisions use the last-confirmed installed linkage: 35 mm carrier, 20 mm crank, 28 mm pickup, 38 mm rod and 55 mm lower leg, with the bare lower-leg part touching the floor. Saved joint references and inversions are preserved. These are V2/P4 changes; the original eight-servo V1 remains a separate supported body.

See what self-leveling is doing#

Body Control now shows measured tilt even while self-leveling is Off, error relative to the saved target, the commanded body correction, and modeled extension and joint-angle changes for each physical leg. Reduced correction is also visible. Per-leg values are commanded/model data, not sensed joint positions or confirmed foot contact.

Control or readout	What it means
Measured roll and pitch	Current orientation from the IMU
Target offset	Difference from the saved attitude reference
Commanded correction	The adjustment requested by the controller
Per-leg changes	Modeled extension and joint-angle changes relative to the nominal motion
Use current pose as level reference	Copy measured roll/pitch into the target fields
Save on robot	Persist the selected targets and existing response settings
Calibrate stationary gyro bias	Collect 256 fresh raw samples, save their mean bias and apply it without rebooting

Capturing a level reference does not change sensor axes or gyro calibration; saving is explicit. Gyro calibration requires a still robot and reports sample variation and storage errors. It does not calibrate accelerometer scale or measure the physical direction of a servo.

Enabled and active remain different states. The recorded Stop command clears the standing balance hold, so leveling can be On while inactive. Stand re-enters the supported hold. A supported chassis may also keep the same measured tilt despite leg correction, allowing the integrator to approach its modeled reach. That explains why commanded adjustment alone cannot prove successful leveling.

The 0.8.48 installation and desktop/mobile controls were checked in the source journal. Physical correction direction and loaded disturbance rejection remain open; this website update did not change the robot’s settings.

Walk: longer steps and a broader lift#

Walk retains staggered footfalls and its existing clock. The foot’s swing now uses a broader arch, clearing the floor earlier at the same peak height, while the support transfer is retimed around the moving leg.

Commanded Walk parameter	Previous	Revised
Full planted sweep	32 mm	37 mm
Low/full-gain lift	4 / 8 mm	6 / 10 mm
Stance fraction	80%	78%
Lateral support shift	9 mm	7 mm
Cycle at Speed 100	0.9 s	0.9 s
Planned travel at Speed 100	44.44 mm/s	52.71 mm/s

The planned travel increase is 18.6%. It is not a measured increase in ground speed. The sampled straight-path study has only 0.064 mm minimum modeled dynamic support margin at one speed, with actual center of mass and friction unmeasured. This is a candidate trajectory, not robust stability qualification.

The PCA9685 driver also staggers the start phases of servo pulses while retaining their widths and one frame transaction. Electrical benefit still needs current measurement and an oscilloscope check; distributing pulse edges does not by itself prove reduced motor-current peaks.

Run, Crawl and Crab#

Run keeps its front/rear paired bound and flight intervals. Crawl and Crab now use the installed bare-leg contact surface instead of the historical boot/toe profile. Their revised swing and body shifts transfer modeled support while preserving planted anchors and the gait clock.

Commanded parameter	Previous	Revised
Run planted stroke	40 mm	44 mm
Run lift range	2–8 mm	5–12 mm
Crawl planted stroke	18 mm	32 mm
Crawl lift range	2–4 mm	5–8 mm
Crab fore/aft stroke	36 mm	40 mm
Crab sideways stroke	24 mm	32 mm
Crab stance width	190 mm	160 mm
Crab lift range	6–12 mm	8–14 mm

At full gain, planned Crawl cycle travel increases about 59.5%; Crab sideways travel increases 25%. Stroke, stance fraction and cycle time all contribute to travel, so these are not measured speed claims.

The support study assumes a centered mass distribution. Crawl’s sampled support margin stays positive in the studied cases; Crab still has negative margins during full turns and a small negative margin in one backward case. The three-foot support model does not qualify the dynamics of Run’s paired contacts and flight. Friction, actual mass distribution, collision clearance, motor load and tracking remain to measure.

A quicker, larger seated Wave#

Wave keeps its seated character, lifts the arm promptly and makes three larger waves before replacing the foot and returning to standing.

Authored timing at 1×	Previous	Revised
Seated entry complete	3.0 s	1.2 s
Arm lift complete	4.2 s	1.6 s
Shoulder amplitude	±18°	±42°
Standing recovery complete	15.2 s	6.0 s
Clip duration	16.0 s	6.2 s

Wave has its own bare-leg contact geometry; the separate Sit command is unchanged. The other three legs and body pose stay constant during the waving portion. Offline checks show positive assumed support margin with the arm airborne, but the inherited seating/recovery transition still has a negative modeled margin. That transition, the larger swing and whole-body clearance need physical testing.

The journal records 51 native tests, 24 compiler/handoff checks and a successful firmware build for Wave, followed by verified installation of 0.8.52. No Wave command was sent during that installation record. The website’s older Blender images and showcase remain dated; they do not depict this revised motion.

Goals above the gait controller#

These body trajectories supply movement to MetaHuman’s new locomotion goal interface. That interface adds distance, turn, approach, follow and visual endpoints with actual progress evidence. The intervening 0.8.51 update installed signed lateral control, retained by 0.8.52. Calibrated navigation and physical arrival accuracy remain unqualified.

Evidence and next work#

The gait journal is the engineering owner for these dated comparisons. This report uses the reviewed local October 15 snapshot, whose changes may precede the repository’s published branch. Its hash and supporting source files are listed in the snapshot manifest.

Firmware installation, mathematical reach and passing software checks do not establish loaded movement quality. The next evidence needed is actual foot clearance, body support, displacement, current demand and response to tilt. Earlier experiments remain in the continuing gait history.