CUBEFLYCONNECTOME RESEARCH
PHYSICS FEED PAUSED
—viewers unavailable
thoughtcrime frog avatar@thoughtcrime___

EXPERIMENT 01 · CONTACT BEFORE CUBE SOLVING

Connecting brain and body

Milestone library ↗

Live measurements.
Contact → face turn → cube solve

01 · Neural contact0 / 3 verified cycles
→
02 · 2×2 cubeAppears after contact gate
→
03 · 3×3 cubeAppears after a verified 2×2 solve
WAITING FOR RUNNER
NEUROMECHFLY PHYSICAL BODYLEFT FRONT LEG
CURRENT EXPERIMENTBuilding the physical experimentThe scene updates when the body simulator connects.
TargetPassive hinged panelFirst verify repeatable neural contact
Panel angle now—Last saved physical state
Peak this cycle—Brief motion, not a completed cube turn
Learning statusNot training yetWaiting for the neural adapter
What is it doing, and what comes next?

The fly has six legs, with jointed segments at their ends. Initial calibration used a hinged contact panel before introducing a 2×2 cube. The current learning assay tests contact with that cube. A learned face-turn controller and full puzzle-solving behavior remain future work.

The neural adapter maps measured motor-neuron activity to one foreleg joint. Imposed stimulation is used for calibration, and stale input returns the joint to neutral. The contact assay tests reward-driven changes in existing memory connections, using fixed sensory and leg adapters. Frozen evaluations compare contact success on reserved target positions. Connecting a connectome alone does not establish learning. The cube uses an idealized fixture with switched constraints and a small declared alignment detent. A turn must physically reach within 0.5° of 90°, have fly contact, and produce the correct measured sticker state. A process restart or loss of neural input resets the current scramble; the completed curriculum stage is retained. Cube solving remains a research goal.

  1. Contact calibrationCurrent test: measure physical contact and panel motion.
  2. Neural controlConnect measured simulated motor output to the leg.
  3. Learn repeatable turnsCompare training against fixed controls on new target positions.
  4. Manipulate a cubeBuild and test the full physical puzzle task.
Earlier artificial cube learnerSaved results, celebrations and downloadable clips

These are results from the earlier artificial controller. They do not show the fly brain solving a cube.

0Prototype steps
0Prototype solves

PROOF, NOT JUST CONFETTI

Artificial prototype: learning results.

Awaiting tests
Before training: —0 checkpoints · same test states
Median successful search—
Median successful moves—

Compare at the same depth. Compute time excludes animation. Short practice scrambles are different from competition scrambles.

CELEBRATION TARGETS
✦ First solveϟ Same-scramble speed record◆ 100ms / 50ms / 10ms / 5ms / 1ms✺ Human-time target

The human 3×3 single record is 2.76s, Teodor Zajder (checked September 12, 2026). A special target celebration is reserved for 20-move scrambles. Computer search time is not an official WCA or robotics record.

THE HIGHLIGHT REEL

Archived artificial-controller replays.

Download first prototype MP4 ↓

Clips include the frog, X mark, handle, difficulty, and compute time. Beginning replay uses saved initial tests, then the first learned solve.

The archive opens with the first verified solve. Nothing is staged.