Mechanics and motion capability#

The V1 asset count is a finite library. A generated V2 program can combine targets, holds and face changes, with actual events requesting recovery. Counting named clips no longer describes its whole motion system.

Movement ranges: compare the quantities first#

Dataset date: September 24. The numbers below are preserved for comparison; they have not been regenerated for the later controller revisions or the October 5–6 compact-linkage candidate. The latest motion entry compares that candidate with the shared model and records its changed poses.

A commanded angle is not a measured mechanical limit. V1 uses logical controller angles. V2 uses signed Blender angles for three coupled inputs. Their zeros, joint axes and mechanisms differ; subtracting the numbers does not give a meaningful percentage increase in reachable workspace.

The September 24 V2 envelope above was extracted from 23 declared gesture sources, 14 locomotion recordings and the Run recording, pooling all four legs by input type. It excludes the retired fixed-angle turn recordings. Its shoulder maximum therefore differs from older audits that included those turns. The forward-walk row includes its recorded Speed changes, advanced settings and Finish; it is not the extrema of every possible walking control setting.

Download the per-recording, per-joint CSV and its read-only extraction script. The CSV includes source hashes, preserves the two angle conventions and also lists all V1 explicit target ranges. The gait journal retains the original V1 channel table and earlier linkage studies.

Controller and processing#

Processor families and board interfaces follow Espressif’s S3 description and the Waveshare board documentation. Memory and clock values above also follow the local board records and checked sdkconfig files. The S3 silicon supports up to 240 MHz; this checkout selects 160 MHz. These clock rates and memory capacities are not an application benchmark or evidence of longer battery life.

The P4 solves the leg linkage directly and plays compact gesture curves through one output controller. The September 28 website’s deployment record reports a 2,370,848-byte application, verified flash and boots, retained twelve-joint calibration and saved motion speed after reboot. Hardware checks cover startup Home and settings; loaded twelve-servo timing, tracking and power remain separate measurements.

The Q6A figures come from the September 28 resource audit. More RAM and different processors allow different workloads, but their specifications are not a measured speedup over V1. Live IMU telemetry and saved leveling controls now have dated firmware installation records. Loaded leveling, the ROS adapter and native-USB body path remain unqualified. SD video and Wi-Fi maintenance are additional P4 features, with controller tests separated from loaded motion.

Power supply, battery, charging and running time#

The separate-cell power arrangement is now identified in the October 26 hardware record. Earlier Anker and SHARGE bank plans remain historical iterations.

The revised V2 battery capacity is unspecified, so there is no current capacity-ratio comparison. A useful runtime estimate requires usable delivered energy in watt-hours and average whole-robot power: runtime in hours ≈ delivered Wh ÷ average W. Idle, walking and expressive motions need separately identified test conditions.

The September 24 Anker study and September 28 SHARGE budget preserve the earlier removable-bank iterations, including the proposed two-port 5 V / 3 A branch allocation. That allocation is historical, not a specification for the new batteries and charging boards. The October 6 rebuild records why the bank was dropped. The October 26 packaging record identifies the later pack, converter, charger and printed parts.

Sources and next comparison#

The motion-range dataset and original V1 comparison use the local Ainekio checkout after commit 3f872e7, including uncommitted September 24 work. The declared catalog, locomotion README, CONTROLLER_VALIDATION.md, model geometry, V1 seed motions, hardware records and both firmware configurations supplied the figures. The source manifest records snapshot hashes. The September 28 changes use robot commit 270537f, current controller validation and saved Blender review notes, identified in the September 28 source manifest. These are dated source and validation records, not a new hardware test performed for the website.

The October 6 manifest adds robot commit 13fc3e4, the later remote status change, local linkage research and Blender review records. The physical standing result, new battery direction and approximate improvement estimates come from Greg’s October 6 report.

The next comparison should add the completed assembly’s mass, physically measured foot workspace, loaded motion observations, delivered battery energy and timed runtime under stated gait settings.

October 8 update: current V2 physical observations and photograph come from the owner. See the new build entry and source manifest. Earlier range datasets and improvement estimates retain their original evidence boundaries.

October 12: shell/power and motion revisions, YOLOE and trained MetaHuman specialists extend this record. Model timings are desktop evidence, not a V1/V2 or Q6A hardware benchmark.