A new finite-event model replaces the continuation loop. Simulated event corrections finish; a separate supported stretch passes physical review, while generic calibration and real recovery remain open.

Compact motion programs, measured shaft arrival and reviewed gesture corrections lead into a 1,261-example Stand training run. Reference agreement improves, but acquiring Stand and finishing compound movements remain difficult.

Current mass, power components, linkage dimensions and support-stand files, alongside the supported Stand/Walk trials and fresh-feedback arrival changes.

Full Auto can continue recorded work and use its outcomes. Live testing exposes the remaining initiative gap; queue, event publication and on-demand model changes improve the supporting runtime.

Physical trials demonstrate modest tracking gains, followed by a host learning loop that records normal walking and compares candidates. Camera estimates, transport gaps and unresolved locomotion quality remain explicit.

A repaired ground connection restores useful joint feedback and steady measured standing in one trial. New face geometry, final-pass verification and connection-loss handling accompany the rebuilt body; inactivity rest is prepared.

A dedicated tip/shake workflow shortens reaction planning. Activated lifecycle and transport repairs finish admitted event work and preserve it across reconnect, while new offline input is not queued for later delivery.

Fourth-wire servo readings now reach the P4 through a CD74HC4067 multiplexer. All twelve loose servos are calibrated; revised chassis parts accompany the rebuild. Loaded standing and walking remain to qualify.

New trained specialists, a separate autonomy adapter and activated execution repairs accompany camera and memory tests. The report separates deployed changes from remaining decision, recall and physical qualification work.

Persona becomes the outward speech owner. Desire review gains standing goals, memory provenance and task results; workflow recovery retains events and admitted graph definitions. Grounded companion behavior remains under qualification.

Selected locomotion goals now connect to measurement and Finish evidence. Signed lateral control has an installation record; camera calibration, model qualification and physical arrival accuracy remain open.

Longer commanded steps, revised bare-leg contact paths, staggered servo pulses and a quicker Wave accompany level-reference and per-leg diagnostics. Firmware installation through 0.8.52 is recorded; loaded performance remains unverified.

Small autonomy and persona fits completed, but actual serving tests retained objective loss and invented experiences. The production setup was restored; revised Task references now follow the single Persona speech contract.

A dedicated Event Trigger workflow receives body measurements and selects an intention for Environment Mode. Robot Status includes IMU evidence; reaction wording comes from the model. Full wake/stretch reliability remains open.

Boot posture is separate from calibration Home. Body Control gains saved leveling strength, damping and attitude targets; the geometry calculator explores load and reach. Installation is recorded, with loaded correction still unverified.

Shared JPEGs feed internal-card video; Body Control retrieves clips and logs and stages P4 updates for explicit activation. A 469-second controller run recorded 411 frames without reconnecting.

Training addresses quoted speech mistaken for physical movement. The latest candidate improves overall reference agreement but misses three historical actions; the existing Task adapter remains in production.

Exterior printing has begun, with more weight modifications and a switch to a 4 A supply. See the current Blender design, explore dated shell and camera-mount files, and compare revised Lay Down and Play Dead paths.

Real camera observations now reach MetaHuman with query and track IDs. The stationary chair test shows both a weak model-selected region and a successful tighter reference, with sustained camera recovery still unresolved.

Follow the trained intent/task split, recorded activations and computer-task delegation. Side-by-side results explain why a later goal-planning candidate was rejected despite a better evaluation score.

Lay Down and Play Dead gain staged foot placements and a reverse recovery path in local source. The journal preserves the original end poses and distinguishes sampled geometry from physical qualification.

The robot can now stand, walk and respond to commands. The exposed assembly, October 6 comparison, and latest mounting-part exports. Active Operator has not been tested.

The physical display is illuminated, and the local expression library keeps growing. Explore actual animated renderer previews alongside the recorded camera, speaker and wake-capture results, with newer local refinements clearly identified.

Request routing and connection changes, plus speech and person-detection work. Model stop/completion failures and physical Active Operator testing remain open.

The latest entry records standing, walking and command response while preserving the failed standing attempt and all earlier linkage experiments. Exact installed revisions and measured gait performance remain to document.

The printed chassis could not stand on its MG90 servos. Follow the material removal, move to separate batteries and onboard charging boards, and assembly changes while replacement parts are printing. The revised assembly remains untested.

Compare the shared model and compact candidate, see which foot placements changed, and follow all 23 remapped motions in Blender. Reduced Upright reach and the provisional servo-travel mismatch remain explicit.

The software now includes steering updates, an IMU estimator and expanded camera/recognition interfaces. Desktop ROS checks and a real hotspot connection are recorded separately from live balance control and sustained wireless reliability still to establish.

The newest journal entry connects the failed standing attempt to the revised linkage, records the complete offline library and separates modeled motion from the next physical test.

Revised v2 geometry, Q6A integration, fresh-frame processing, and saved MetaHuman executions. YOLO and the connected active loop are still being worked out.

The project now has two build directions: onboard Q6A with a planned USB body link, or offboard Q6A over Wi-Fi. Compare their packaging and power budgets, the installed ROS 2 environment, and the remaining work toward responsive gait.

Actual Blender images follow the new build scenes, slimmer tray, LCD carrier and alignment marks. The latest front/rear clearance study keeps the original front carriers and moves those legs forward; body refitting remains in progress.

Before/after tables explain the Walk, Run and Crab changes and distinguish continuous gait controls from saved named-motion speed. Offline geometry and settings checks are recorded separately from the local IMU correction loop still to implement.

The comparison now includes Wired and Wireless V2, the new bank and current control changes, while retaining the explicitly dated September 24 motion-range dataset and unmeasured runtime.

The red body gains compact mask mounts and reinforced tabs while the underside is fitted around the Anker bank already on hand. Follow the width changes, battery clearance studies and remaining assembly work, with actual Blender images.

Compare the physical V1 and evolving V2, with fresh per-joint range data, ESP32-S3/P4 configuration, the two power arrangements and an explicit record that neither runtime has been measured.

A playable Blender showcase accompanies the new Speed, stride, cadence and Finish controls. The latest entry records revised expressions, six-direction Crab and the September 24 application flash, with the physical tests still ahead.

The walking sequence lengthens its steps, doubles its cadence, then returns to short steps. Revised linkage, stance planning, and motion data.

Actual assembly views follow the face, collar, roof and fin work, including rejected trials and a matched roof before/after comparison.

Earlier and compact legs compared side by side, with the horn-cavity correction and upright print files.

The narrow-body revision moves the side assemblies inward and revisits electronics packaging. A later front-panel change moves the camera-cable recess below the face.

Dated entries add the motion library and 10/20/30 mm compact-leg probes, with comparison tables, contact assumptions and downloadable results.

A continuing technical journal compares the original robot's stored motion ranges with the new linkage's theoretical travel. The first entry includes the mechanism diagram, nonlinear input/output examples, and downloadable range data. Gait calculations are still ahead.

The printed legs have much more expressive movement than the models I have seen online. Three prototype photos, fit problems, revised dimensions, and foot options.

The four-leg Blender assembly now includes a wider chassis, electronics, smoother linkages, revised covers, and a camera/display carrier. Fourteen saved revisions, component dimensions, remaining fit checks, and a MetaHuman OS update.

A direct rod, a supported relay yoke, and an offset cascade were compared. The relay remains the next research candidate; none met the complete range and pose brief.

Robot tasks now follow a finite plan with a clear completion condition. Pending ideas remain visible without becoming standing movement instructions. Deployment verification remains.

Delayed feedback stays attached to the original command. Recovery tracks the result without repeating movement. Controlled tests passed; a gateway restart and live testing remain.

A live review found command feedback, images reaching model requests, and acknowledged speech deliveries. It also exposed stale goals. Later repairs retain truthful execution status and connect specialist answers to the waiting request; physical validation of those repairs remains.

MetaHuman’s development code saves what the robot was asked to do, what has finished, and what it is waiting for. This helps a request continue after an interruption. Focused software tests have passed; deployed robot tests remain.

V2 adds a sideways hip joint to each leg. An OpenHarmony Puppy V2 reference leg will be tested before the chassis and power supply are sized. Prepared movement sequences come first; adaptive balance is deferred.

The existing robot’s parts list now identifies its Freenove controller, display, and 1000 mAh battery, and corrects wiring assignments. These V1 details do not determine the planned V2 wiring or power supply.

A local guide replaced five remote-document pages and earlier fantasy artwork. It covered hardware, firmware, commands, gateway, camera, speech, MetaHuman integration, and autonomy.

Robot Operator became responsible for when the robot could start an activity. Other software selected actions, carried commands, saved progress, and scheduled jobs. September’s redesign later changed how ongoing requests resume.

The pilot sometimes treated ordinary room sounds as the wake phrase. Its detection threshold and test recordings needed more work; an accepted production wake model is still outstanding.

Firmware gained the corrected OV3660 capture profile and more named motions. The simulator was also updated at the time; that project is now retired.

Prepared walk sequences were shortened to roughly five seconds and installed in the robot’s LittleFS asset storage. Reading them back and comparing digests checked the installed bytes; it did not measure gait quality.

Pacing audio delivery produced clean audible speech in one test. Logs still showed the speaker queue running empty and faults sending microphone audio, so repeated conversation needed further work.