650 g, revised power packaging and a training stand

October 21 workbench photograph.

The October 26 hardware record puts the assembled robot at 650 g and confirms that the physical build matches the current Blender design. The earlier 566 g figure belongs to the October 13 assembly. SHARGE is no longer used; the current power assembly has a 2S battery, Mini560 buck converter and HiLetgo 2S USB-C charger.

Parts and packaging#

Part or dimension	Current record
Battery pack envelope	73.6 × 46 × 22 mm
Mini560 converter	30 × 18 × 5.6 mm
HiLetgo charger	36.5 × 17.75 × 1.5 mm bare PCB; 6.8 mm modeled overall height
Speaker	20 × 30 × 6.66 mm
Carrier / crank / pickup / rod / lower leg	35 / 20 / 28 / 38 / 55 mm
Front-to-rear shoulder spacing	173.434 mm
Left-to-right shoulder spacing	63.500 mm
Joint feedback	Twelve modified Miuzei MG90S servos through a CD74HC4067 multiplexer

The current Blender file is Ainekio-Motion-Working.blend, scene Wireless, with one Blender unit representing one millimetre. The October 26 frame-168 measurement spans 245.85 × 182.79 × 109.06 mm. That is the outline of that raised-leg pose, not the normal standing footprint.

The power record includes an 8.4 V battery and 5 A buck rating from October 15, a separate 5 V servo rail, common ground and 3.3 V logic. Current pack capacity, loaded current and running time have not been measured. The older power-bank energy and runtime estimates do not describe this pack. The Q6A remains offboard and has its own power.

A new one-piece rear power-carrier STL joins the earlier open-frame and separate retainer exports. The new file is included below; its filename alone does not establish which carrier revision is fitted.

Compact support for motion training#

The new fixed training stand uses three printed parts: base, ribbed upright and vent-locating cradle. It replaces the earlier tall tilting/clamping design. The robot rests on the cradle and lifts straight off.

Stand detail	Dimension or hardware
Base footprint	140 × 110 mm
Overall height	77.1 mm
Tray support height	75.9 mm above the base bottom
Locating blades	Two, each 4.0 × 40.0 mm; 42 mm between centers
Matching modeled vent slots	4.4 × 40.4 mm; nominal 0.2 mm clearance per side
Blade engagement	1.2 mm, with tapered lead-in
Base-to-upright fasteners	Two M3×20 socket bolts and two M3 nuts
Cradle-to-upright fasteners	Two M3×10 socket bolts and two M3 nuts
Solid model volume	58.0 cm³, down from 440.8 cm³; approximately 87% less

The cradle shoulders carry the tray; the locating blades do not extend into the battery. Base and cradle print flat underneath, and the upright prints on its broad side. Insert the upper nuts through the upright’s side pockets before attaching the cradle.

The volume comparison is geometry, not a slicer filament estimate. Current-pose clearance and mesh topology were checked in Blender. Printed fit, full motion clearance and physical loading still need testing. The stand supports the body; it does not restrain it against an upward movement.

Download the three stand STLs, assembly notes and one-piece power carrier.

Firmware work at the bench#

An installed firmware revision rebooted when starting Stand or Walk. The motion planner was consuming too much of its 16 KiB task stack. Version 0.8.123 moved the candidate planner into task-owned storage; the compiler’s start_motion stack figure fell from 8,496 to 3,760 bytes.

After installation on October 25, the supported robot completed Stand and one finite Walk cycle with its feet clear, without rebooting or disconnecting. Those trials verified the repaired command path. They were not floor walking.

The following revisions address a different problem: claiming arrival before the shaft measurements catch up.

Revision	Change
`0.8.124`	Stand stays pending until fresh feedback verifies its final targets
`0.8.125`	The moving leg must reach each lift/place landmark before the next segment; saved arrival tolerance confirmed at 100 µs after reboot
`0.8.127` source	Boot/wake and inactivity requests go to Freestyle; generated targets use the same feedback executor

Loaded servo chatter remains unresolved. The earlier 2 mm lift-and-return settling command was installed, but I reported no visible settling movement and continued chatter. The 100 µs value is a shaft-arrival setting, not measured foot clearance.

The next physical comparisons are the loaded rise, foot placement and floor walking with the matching motion model and firmware. Freestyle training results explain why a completed model fit is not yet a reliable standing controller.

Records#

This entry follows the October 26 V2 specifications, the training stand’s Assembly.txt, current STL exports and the P4 firmware log. Source revisions and file hashes.