October 8: The lighter robot stands, walks, and responds to commands.

A stronger lifting arrangement without losing the character#

The printed chassis could not stand on the MG90 servos. Alongside removing weight and changing the battery plan, I have been revising the legs in Blender. The aim is more lifting capability and quicker movement from the servos I am keeping. The replacement parts are still printing; the revised robot has not passed a physical standing test.

The October 5–6 work now includes a complete offline candidate for all 23 named motions. That matters because the robot needs to keep its gestures as well as its walking motion. The new linkage changes what those poses require.

Blender render of the compact-linkage robot performing the Point gesture. — October 6 Point preview from the Blender scene “Motion Library - 35-20-28.” The complete modeled body includes appearance parts that have not been printed. This is an offline motion preview.

Geometry, side by side#

These dimensions compare the shared software model at commit 13fc3e4 with the new experimental model. They are model dimensions, not a measurement of every link currently installed on the printed robot.

Dimension	Shared model	October 5–6 candidate	Change
Carrier	40 mm	35 mm	−5 mm
Crank	24 mm	20 mm	−4 mm
Pickup radius	24 mm	28 mm	+4 mm
Connecting rod	38 mm	38 mm	Unchanged
Distal leg	55 mm	55 mm	Unchanged

The candidate is identified as compact-35-20-28-38-55-visible-20261005. It lives in a separate development package and Blender scene. The handoff explicitly records that the active shared model was not replaced and this candidate was not installed in firmware.

My estimate of approximately 20% more lifting force is a design estimate, not a measured result from these motion checks. The carrier reduction itself is 12.5%; it should not be relabeled as a measured 20% change in either travel or force. The whole linkage and its changing poses still need to be tested.

Preserving the motions, recording the compromises#

Thirteen clips preserve their original body trajectory, foot positions in the horizontal plane and full-sole height: Celebrate, Curious, Cute, Dance, Freaky, Lay Down, Sad, Shake, Shrug, Stretch, Surprised, Swim and Worm. Crouch uses the native inverse-kinematics planner with the same targets, including its −35 mm terminal body translation.

Other movements needed explicit changes:

Motion	Retained behavior	Change in the compact candidate
Sit	Body trajectory, timing, contact sequence and supporting-sole orientation	Rear landings move about 5.584 mm forward
Wave	Seated three-wave sequence and return to standing	Uses the revised Sit
Upright	Sit entry, two front steps and full −90° body rise	Refitted body path; selected forelimb reach falls from 90.339 to 76.585 mm
Rest	Step timing, final boot orientations and chassis lowering	Landings move about 5.018 mm forward
Bow / Nod	Foot paths and original timing	Body translation adjusted by as much as 5 mm
Pushup	Five repetitions and timing	Body translation refitted by up to 4.767 mm
Play Dead	Remaining sequence and timing	Front slide endpoint moves about 7.158 mm rearward
Point	Original carrier angle, boot orientation and extension timing	Pointing boot displaced by 5 mm because the carrier is shorter

Blender view of the compact-linkage Sit study with the rear legs folded. — October 5 seated-support study. The rear foot-placement adjustment lets the new linkage follow the intended sequence; it does not establish physical balance or load capacity.

Upright is being kept as an experimental, best-effort motion. Its 76.585 mm reach describes this selected fit, not the maximum reach of the mechanism in every direction. The original extension would require a different linkage branch, and the retained fit reaches an input dead center. On October 6 I chose to continue with the rest of the library while keeping these limitations visible.

Blender preview of the experimental Upright pose on the compact linkage. — Experimental Upright study from October 5. Reduced reach, the dead-center finding and unqualified physical feasibility remain part of the record.

What the checks establish#

The final library contains 31,031 source/preview samples. The recorded native sampler checks cover all twelve joints at 62,517 timestamps. Forty-six nonsequential Blender poses agree with the native forward-kinematics calculation to within approximately 0.000027 mm. The saved scene was reopened and checked.

Those results establish agreement between the motion data, software calculation and Blender scene. They do not establish that MG90 servos can carry the load, that the body avoids every collision, or that the robot balances.

The native test record is six passes out of seven. v2_upright still fails its historical requirement for more than 90 mm forelimb reach. That assertion has been retained as comparison evidence; the reduced-reach experiment is not being described as a clean test pass.

There is also a servo-travel problem to resolve. Across Upright and Bow, the front cranks require approximately 243.868° of combined travel. The provisional mapping covers 225° over its 400–2,900 μs reference interval. Simply shifting the center cannot fit that span. Actual usable shaft travel has not been measured; these reference values are not installation instructions.

Continuous gait work remains connected to this study#

The earlier continuous-motion candidate checked 70 planner cases, retaining the intended foot targets, phase/contact schedule and shared motion clock. The October 6 library then completed the named gestures that were still missing from that earlier handoff.

Separately, the committed controller now uses a 96 mm full Walk sweep, 8.85 mm Walk lift and a −8 mm body offset. Continuous Crawl uses a −22 mm body offset, an 18 mm sweep and 4 mm lift. Crawl and the named Crouch gesture are different motions; their offsets should not be interchanged. A shared joint speed limit can stretch automatic motion timing while preserving the planned path. Requested cadence is therefore not proof of achievable loaded speed.

The next evidence needed is the revised printed assembly: actual mass, usable servo travel, standing ability and movement under load. That will determine which of these modeled gains survive the physical build.

Record behind this entry#

The committed model and controller notes provide the baseline. The local October 5–6 records are docs/gait-research/linkage-motion-parity-20261005/handoff.json and docs/gait-research/seated-support-20261005/, particularly its README, library-handoff.json, library-native-check.txt and library-travel.json. These experimental files were local development artifacts at review time.

The review manifest records their hashes and image origins. The dimension comparison CSV contains the five dimensions above. This entry continues the gait development journal.