October 27 motion follow-up#

A supported stretch verified all nine targets after calibration positioning. The asymmetric compensation from that trial is kept separate from the generic training motion. Freestyle now uses complete programs and actual body events for recovery; loaded performance remains open.

Current hardware: October 26#

The current body record reports 650 g, twelve feedback-modified Miuzei MG90S servos and a 2S battery / Mini560 buck / HiLetgo USB-C charger assembly. SHARGE is no longer used. Pack capacity and running time remain unmeasured. The compact linkage lengths are 35 / 20 / 28 / 38 / 55 mm for carrier, crank, pickup, rod and lower leg.

The new fixed training stand has three printable parts and a 140 × 110 mm base. Download its STLs and the one-piece power carrier. Supported Stand and one finite Walk cycle completed after a firmware stack repair; loaded chatter and reliable floor walking remain open. Freestyle’s new target and training work uses shaft arrival to work toward those results. The October 27 follow-up replaces the continuation loop with finite programs and event-triggered recovery.

October 21 revision#

New photographs and body-control results show the reassembled robot, revised red face surround and brighter LCD design. A ground repair restored useful feedback and a measured Stand-and-hold trial passed. Installed firmware is recorded through 0.8.113; physical disconnect behavior and the newer inactivity-rest feature remain to test.

Loaded gait trials and host learning now connect joint, IMU and camera observations to separate Walk/Run policy comparisons. Earlier tracking gains are modest; reliable locomotion and metric camera accuracy remain open. See the October 21 build.

October 18 feedback rebuild#

The feedback-servo rebuild adds fourth-wire position readings from all twelve servos through a CD74HC4067 multiplexer. Unloaded calibration is complete; measured Stand remains disabled pending supported tests with the linkages reinstalled. Walking trials still showed sagging and foot drag. Revised chassis exports include battery, tray, servo, gyro and display mounts. See the physical rebuild in the gallery.

October 15 motion revision#

Revised motion and leveling controls cover Walk, Run, Crawl, Crab, a larger seated Wave, staggered servo pulse phases, a level-reference control and stationary gyro calibration. The journal records installation through 0.8.52. Commanded leg diagnostics are not measured joint positions; loaded gait, gesture clearance and leveling performance remain open.

October 14 baseline#

The hardware journal records 566 g, 5 V and MG90 servos for the October 13 assembly. The startup and leveling report covers staged boot posture, saved self-leveling controls and the new local geometry calculator. Firmware 0.8.46 installation and live control readback are recorded; loaded self-leveling remains unverified.

Body Control also gains body-event settings, SD recording, downloads and Wi-Fi maintenance. The full model-selected wake/stretch routine remains incomplete. The existing thumbnail is the October 12 Blender view, not a new assembly photograph.

Shell and power report: October 12#

I have begun printing the shell, made further weight modifications, and switched to a 4 A power supply. The later October 13 report specifies 5 V; measured current draw and the final battery configuration still need recording. The new build entry contains current Blender views, 27 prototype exports and revised grounded-motion comparisons. Shell fit and Active Operator testing are still pending.

Physical movement baseline: October 8#

The robot can stand, walk and respond to commands. This follows the failed standing attempt two days earlier. The carapace is not attached and Active Operator has not been tested. The physical build update includes the current photograph and a side-by-side account of the two attempts.

The latest committed exports add LCD-carrier braces, a 20×30 speaker holder, compact carriers/cranks/pickups, and electronics/power mounting parts. Local camera-carrier and longer front-tray exports are still being developed. The October 14 firmware record and October 13 mass/voltage report add evidence; a complete reproducible assembly record remains open. Earlier 20% improvements remain estimates.

The October 6 power decision dropped the heavy 10,000 mAh rechargeable bank in favor of separate batteries and onboard charging circuit boards. The October 26 entry above records the selected boards and pack arrangement. Cell capacity and runtime remain unmeasured. The earlier Anker and SHARGE studies remain dated design iterations.

Wireless, with the Q6A offboard over Wi-Fi, is the current build priority. Wired retains the onboard-Q6A/native-USB alternative. The October 5–6 offline leg candidate uses a 35 mm carrier, 20 mm crank, 28 mm pickup, 38 mm rod and 55 mm distal leg. All 23 named motions have been remapped in Blender, with reduced Upright reach and unresolved servo travel. The installed walking build’s firmware/model revision still needs recording.

The P4/gateway source now includes continuous steering and shared joint-speed control. IMU acquisition and configurable balance correction are now connected. Loaded correction remains unverified. ROS 2 tools have recorded checks; the robot adapter and native-USB body transport remain integration work. The October 13 Wi-Fi recovery run passed for 469 seconds; earlier sustained-session failures remain part of the connection history. V1 remains supported. V2 stood and walked on October 8.

The September 15 prototype post contains physical leg photographs; the V1 hardware guide records the earlier robot’s parts and wiring.

Changes from V1#

September 7 design comparison. The current comparison is linked above.

September 7 concept art: V1 is on the left and an earlier V2 concept is on the right. The leg reference is OpenHarmony Puppy by 杜邦线丶小战士 (pcbguy). The final geometry will follow leg tests and component measurements.

MetaHuman OS, the software that handles conversation and decisions, will continue to request actions such as “stand” or “walk” through the robot’s network gateway. The controller on the body will turn those requests into joint movements and enforce their limits.

Legs and chassis#

Earlier design direction, retained for the build history.

The early V2 leg plan had hip roll for sideways movement, hip pitch for forward and backward movement, and a knee that bends the leg. Together they provide more foot-placement options and room for body lean and expressive poses. Speed is not the first priority.

OpenHarmony Puppy V2 provides a reference for lightweight MG90S legs, supported pivots, and linkages that keep motor weight close to the body. Local files include printable STL models and STEP assemblies converted for Blender. Testing will establish how those ideas fit Ainekio’s shorter, broader shape.

The chassis is being assembled in Blender around the leg mounts, board models, connectors, and front camera/display carrier. The original OpenHarmony-derived assembly has since been revised through the direct-rod and compact-leg work. The September 9 short-rod study remains a separate historical comparison of alternatives.

Cables need clearance throughout a leg’s movement. Programming, charging, power connections, legs, and electronics should remain accessible for replacement and repair without dismantling the whole body.

Printed leg development: September 12–14#

The new prototype photographs show a printed leg assembled in partly folded, folded, and extended poses. The records now include a failed shell fit and horn measurements from actual hardware, as well as saved Blender changes. Many refinements apply to one front-left leg or a separate review file; they are not all combined into one released four-leg print set.

The shoulder-motion work is still an unfinished chassis-clearance study. Its 270° swivel-servo brief does not establish usable assembled travel: the recorded draft has pose-dependent interference and awaits final review. The latest build log shows the source images and explains the alternatives and their remaining limitations.

Parts in the September 10 Blender assembly#

This is the earlier electronics/body snapshot. The leg refinements above are newer; inclusion here does not make a reference board or display a final choice.

These quantities describe models in the assembly or the stated servo plan. They are not a purchase checklist. Dimensions below are in millimeters; board dimensions exclude connectors unless stated. Mounting patterns are distances between hole centers. The electronics import notes and front-face design record identify the following references.

The P4 model came from the supplied ESP32-P4-WIFI6_3d-20260109.stp file. Its four polygonal mounting openings measure about 1.66 mm across the inscribed circle, on a 54.164 × 18.25 mm pattern. These are model measurements, not a manufacturer drilling specification.

The Q6A and PCA9685 models came from Radxa’s hardware downloads and Adafruit’s CAD files. The display uses the Waveshare module reference and its mechanical drawing. The camera board dimensions come from the Arducam B0033 drawing.

How the Blender assembly was built#

The revision names below refer to saved Complete-WIP Blender files. Their order documents the modeling work, not a completed physical assembly procedure.

WIP 5 board placement, before the later covers and face. This records that revision’s layout; it is not a wiring diagram.

Robot Layout → LeftRightSpacing_mm controls the leg roots, center mounts, inserted panel sections, covers, and later face mounting positions. At the 97 mm setting, the covers are 112 mm wide. Widening moves the existing mounting features rather than scaling their holes. The cover and face checks also exercised the 107 mm setting.

The WIP 5 layout provides a 6 mm gap behind the Q6A tray and between the rear board envelopes. Four tray cutouts include 0.3 mm around modeled connectors. Those clearances belong to that saved layout. Real plugs, cable bends, cooling, and full leg motion need a separate check; the tray still overlaps the later panel and bottom cover and requires trimming.

Choices and alternatives#

Exact fastener counts and lengths, bearing/shaft procurement, filament and print settings, V2 audio hardware, battery, regulator, harness, and assembly torque are not settled in the reviewed records. This page will need those measured details before it can serve as a complete replication guide.

Controller and motion: earlier plan and September progress#

The paragraphs below preserve the earlier controller plan. By September 17, the project had a separate twelve-joint model and thirty geometric motions, and the selected signal driver was a HUAREW PCA9685 breakout. That hardware selection is distinct from the Adafruit board model in the September 10 Blender table. The integration record reports a built motion image, pending flash and unavailable physical motion.

ESP32-S3 and ESP32-P4 name different controller chip families. Choosing the P4 does not make the existing firmware ready for the new body. Board pins, communication hardware, camera, and audio still need their own implementation and tests; V1’s S3 wiring map is not a P4 wiring guide.

The original plan favored twelve servo-control signals generated directly by the P4. The later Blender model includes a PCA9685 pulse-width modulation (PWM) board. Both approaches still require pin allocation, control software, calibration, and peripheral tests; a board model is not evidence that either path is wired.

A scripted motion consists of frames. Each frame specifies a target angle for each joint and how long the transition should take. For V2, a frame will contain twelve targets. Interpolation supplies the positions between frames so the robot moves smoothly instead of jumping from one pose to the next.

Planned controls include center calibration, direction inversion, minimum and maximum angles, velocity and acceleration limits, safe rest poses, and staged startup and stand-up sequences. Existing movements can first be adapted with the four new lateral joints held neutral, then use those joints selectively.

Walk, stop, stand, sit, bow, and wave remain examples of higher-level requests. At the initial planning stage, the body core, motion assets, protocol validators, and MetaHuman motion-plan schema described eight joints. They needed to agree on the new joint names, limits, and advertised capabilities before accepting twelve-joint plans.

Power: measure the load before choosing the supply#

Earlier electrical planning, retained for context. The October 6 direction uses separate batteries and onboard charging boards. The bank option below is historical; the final cells, boards and power arrangement are unspecified.

The owner reports V1 operating on its existing 5 V / 5 A supply. The design note proposes measuring that robot first, then evaluating a 5–6 A supply line, or rail, under V2 loads. Four extra servos mean the V1 result cannot establish the new robot’s requirements.

Controller and servo power will branch near the supply. Motor current will bypass the P4 board, while a common ground gives the controller and servos the same electrical reference. A bulk capacitor near the servo rail can help with brief current demands; the supply still has to sustain the load. Smooth acceleration and staged movement will be part of testing.

A battery or power bank must deliver the needed current at 5 V continuously, cope with startup and motion peaks, and remain stable in voltage and temperature. An advertised USB-PD wattage does not by itself establish those properties. The final power source remains undecided.

Build and test sequence#

1. Make one leg. Measure fit, servo horn alignment, travel, interference, stiffness, bearing support, printed weight, and reachable foot positions. The horn is the arm on the servo shaft; backlash is unwanted joint play. Both affect how accurately a printed assembly follows a commanded angle.
2. Measure V1 power use and adjust the leg design. Use the existing robot and the experimental leg to inform the mechanical and electrical choices.
3. Test the P4 in stages. Check signals and calibration, then one loaded leg, two legs, and all twelve servos.
4. Validate the modeled chassis and design the wiring harness. Check them against measured parts and power needs, with room for motion and service access.
5. Port scripted movements. Tune transitions before the final complete print.
6. Repeat normal movements under load. Test standing, sitting, walking, turning, and stopping. Watch for binding, supply shutdown, brownouts (voltage drops that reset electronics), and frequent near-stall operation. Check controller capacity and repair access as well.

Later options#

The initial plan deferred IMU stabilization and real-time inverse kinematics. The current V2 controller now implements both, including configurable local leveling; successful loaded correction still needs physical validation. Adaptive terrain response, foot-contact sensing and learned locomotion remain later work. Higher-cost servos and wireless charging were outside the initial build plan. Onboard Linux computing was also excluded from that first scope. The Q6A model records the September 10 wide-body packaging study; its board and tray were retired from the September 17 narrow chassis. No onboard Q6A runtime is established by either Blender layout.

The MetaHuman redesign can develop against V1 while these mechanics are tested. Current project status tracks the latest summary. This page’s September 10 mechanical details come from the electronics import notes and the WIP 5, 10–14 design and saved-file reports. Their checks cover saved geometry and static fit; no new physical test was performed for this documentation update.