Fourteen CAD revisions: chassis, electronics, linkages, covers, and a camera/display carrier, with a brief MetaHuman OS refinement update.
Published 7 min read
Most of the last two days’ Ainekio work has been in Blender. The V2 design now has four leg assemblies arranged around a body, a wider electronics bay, and models of the boards that need to fit inside it. Work has also continued on the leg housings, linkage curves, front connector panel, fitted body covers, and a removable camera/display face.
These are CAD models and geometry checks. V1 remains the working physical robot; the twelve-servo V2 still needs printed parts, assembly, wiring, and motion tests. The images in this post are renders of the design files. The cover image shows the WIP 14 camera/display assembly saved at 18:45 PDT. Its face graphic is a visual preview, not running display software. Blue covers in the earlier images are render coloring, not a material selection.
The earlier widened-chassis stage, with the electronics bay open for inspection.
The first assembly revisions brought the four legs together and fitted the body’s center mounts and end panels. The electronics then gave the layout a more concrete set of dimensions to work around.
Three board models are included: a Radxa Dragon Q6A computer, an ESP32-P4-WIFI6 controller board, and an Adafruit PCA9685 servo-driver board. The Q6A and PCA9685 came from manufacturer CAD sources; the P4 model was supplied separately. They were imported at millimeter scale, including the connectors and components that extend beyond the circuit boards.
The chassis was widened by 40 mm—20 mm on each side. An editable spacing control moves the leg roots, center mounts, and inserted middle sections of the front and rear panels together. Existing bearing and screw features retain their shapes as the width changes.
Electronics placement in the widened-chassis revision. The boards are CAD models; this is not an installed or powered assembly.
The Q6A sits at the front with its ports facing the front panel. The other two boards sit beside each other toward the rear. The layout includes a Q6A tray, connector reliefs, and nominal gaps between the board envelopes. Cable plugs, bend radii, cooling space, and the remaining board fixings still need attention. Placing a computer in the model does not establish an onboard MetaHuman installation.
The upper leg housings have gone through rounded-profile and opening-layout revisions. After further manual editing, the chosen Part 3 housing and its editable cutters were synchronized across all four legs. The saved-file report confirms matching evaluated geometry while preserving the assembly placements.
Earlier housing variants recorded reductions in CAD solid volume, but those figures belong to those variants. They are not a measured weight saving for the later manual revision or a printed robot.
The Part 26 linkage also received more clearance and a smoother outline. Its center first moved 3 mm farther outward, then the concentrated bend became a broader curve that blends into the round mounting ends. The two oval slots follow the new outline. The 4 mm plate thickness and functional mounting geometry were retained.
The later curve revision, applied consistently across all four legs. Static interference checks are recorded; full mechanism travel remains to be tested.
The front panel now closes around the Q6A’s two USB stacks and Ethernet jack: three openings serving four USB sockets and one network port. The editable cutters follow the board, and the panel’s width adjustment was checked at two settings.
Front-panel revision from the WIP 11 assembly. The existing tray lip remains visible below the connectors and is still due for trimming.
An internal relief clears the board edge and nearby underside components. The port openings currently match the modeled connector envelopes without extra nominal clearance, so manufacturing tolerance and real cable-plug fit remain open questions. The Q6A tray was deliberately preserved; its front lip still intersects the panel.
WIP 12 added fitted top and bottom covers, each widened from 72 to 112 mm while preserving the original 3 mm plate thickness. Their eight frame mounting holes align with the current chassis, and both covers follow the same adjustable width control as the rest of the body.
The saved checks describe closed solids with round hole boundaries retained and clearance from the board models and servo housings. A small existing height step in the frame leaves 0.2 mm seating gaps at some mounts. The unchanged Q6A tray also overlaps the bottom cover, extending the same tray-trimming job identified at the front panel. Printed fit and strength still need testing.
WIP 12: an earlier cover revision, retained here to show how the design changed.
WIP 13 then closed 18 unused openings in the top cover and 15 in the bottom, including the large access cutouts. Each plate now retains only eight 3 mm frame screw holes. The 112 mm width, 3 mm base thickness, outer clearance contours, and bottom tabs were retained. The Q6A tray still needs trimming.
WIP 14 places a camera above a landscape display on one removable carrier. The LCD is a Waveshare 2inch LCD Module reference, imported from its official STEP model. The OV5647 camera reference uses the Arducam B0033 board and mounting-hole dimensions, with approximate thickness, lens depth, and connector envelopes. Neither reference establishes the final purchased module or its electrical compatibility with the V2 controller.
The carrier uses two existing front screw positions on the top cover, adding 3 mm to their stack. The LCD model includes four M2×6 spacers; the camera rests on four integrated 4 mm standoffs. Final screws, retention hardware, cable routing, and plug clearance still need to be resolved.
Actual CAD render. Rear cable windows are modeled; the wiring harness is not.
The saved-file report records all 259 existing objects preserved and the new face geometry checked at two chassis widths. This is saved CAD evidence; physical fit, optics, cooling, and movement testing remain. The V2 parts and assembly record lists the board dimensions, mounting patterns, and component sources.
A parallel study compared three ways to drive a knee with a stationary servo and short rigid rods. It produced a preferred relay mechanism for further research, along with pose tests and collision findings. None of the candidates met the complete requested movement range and pose brief.
That is useful design evidence, but it is a separate research path. The four-leg assembly shown here should not be read as a validated implementation of that mechanism. The leg mechanism comparison records the rod lengths, measured model ranges, rejected poses, and why the relay variant remains a research testbed.
MetaHuman work has focused on making the existing workflow more consistent. Follow-up input received while robot or specialist work is pending stays with the saved request, and input handoffs check that the execution is compatible with the current software. A successful handoff can report progress without being mistaken for a missing chat response.
Action-selection refinements separate the next action from the broader goal’s progress. Invalid model answers return through the existing correction path, and selected images retain their matching identity and context. Training and profile-memory reset controls have also received clearer lifecycle handling and more focused validation.
The September 10 handoff review records 236 passing durable-execution tests, alongside graph, type, transport, architecture, and isolated-build checks. Training and reset have their own recorded software and synthetic-data checks. These are dated development results, not a new physical Ainekio trial or proof that all current refinements are deployed. The integration guide explains the workflow they are improving.
The CAD assembly makes the remaining mechanical questions easier to inspect: trim the tray and check cover seating, allow for real connectors and cables, check the complete leg movement range, and print representative joints and mounting interfaces. Fit, backlash, strength, power demand, and cooling need measurements on parts and hardware.
This update draws on the September 9–10 Blender revision folders, their READMEs,
saved-file and static-fit reports, and the MetaHuman development ledger.
The mechanical sequence runs from the four-leg layout through the widened
chassis, synchronized housings, smooth linkage, Q6A panel, revised covers, and WIP 14 camera/display carrier. The
MetaHuman checkout was b57d2058 with additional local refinements. Existing
validation records were inspected for this post; mechanical and runtime test
suites were not rerun as part of writing it.