Sibo (@sibochou) profile picture

Funny, positive, creative, innovative

Melbourne, FL
Joined May 2011
Photo by Sibo (@sibochou)Photo by Sibo (@sibochou)Photo by Sibo (@sibochou)Photo by Sibo (@sibochou)Photo by Sibo (@sibochou)Photo by Sibo (@sibochou)Photo by Sibo (@sibochou)Photo by Sibo (@sibochou)Photo by Sibo (@sibochou)Photo by Sibo (@sibochou)
4 electronics boxes on one 2x4 ft tray: a 3+ hr install becomes ~1 hr. smart. then the seats go in first, and two techs can't turn a 2x4 ft tray around in a cockpit. 1 hr in mocap caught it years early. bonus: the recording's expert hands-on motion data robots don't have.
8
90 min to set up, 1 hour to run: a crew escape from a cockpit, years before first flight. furniture comes with instructions. the customer got a video instead, not Just Read the Instructions. (some assembly required)
4
Rescuing someone who passes out at the far end of an aircraft fuel tank: 1 ft tall, 3 ft wide, no room to turn around. Rope on the legs alone didn't work. A crawler system did: 30 min vs 2+ hours to cut the wing skin (simulated in the HIVE). nobody's skin on the line.
14
HIVE job with almost no human factors in it: we placed flight-test cameras in a stealth jet's weapons bay 2 years before the jet existed. one camera showed a missile hitting structure on release. no F-bombs dropped, in both senses. 2 months of camera work became 4 hours.
13
2 inches. That's how short the 95th %ile male fell reaching his own wing's actuator — every human size failed. We moved the access hole 4" in the HIVE. Saved $900K, and skipped cutting a hole in airplane #1's wing.
8
We use six markers. The AI gave him forty. Outtakes from last week's $3M mockup post. New hair, new gloves, markers nobody asked for. A bad take, I shoot again. A $3M jet that came out wrong, you park. Turns out AI still can't do hands. Ours are millimeter-accurate.
8
6 fuel tank sensors passed every engineering check. They failed one test: could she wrap her palm around it. Engineering said leave it. HSI pushed back, leadership sided with the hand. Redesign took a day -- nothing had been built yet.
6
2 hours in sim caught what would've cost 6 months on the line. We simulated wire-bundle routing in a cramped avionics bay before the jet existed. Planned install order would've sealed the tech inside. Re-sequenced it — free. Alternative: ripping out already-installed wiring.
1
5
Same session: we captured exactly how a real technician's hands route and secure wire bundles through a tight portal, high-fidelity, registered to the CAD. Factory paid for design validation. That hand-motion data is exactly what's training humanoid robots to do dexterous work.
6
A customer wanted a toilet in an aircraft that never had one. We simulated the removal before it was ever installed: 10+ major parts have to come out just to carry it back out again. We couldn't fix that. We proved it, timed it, and let the customer decide.
1
1
15
That capture — a real technician performing a real removal and reinstall, registered to the CAD — is also exactly the training data humanoid robots are starving for. The factory pays for the validation. The robot data is a free byproduct.
1
16
Bruce Lee's one-inch punch got its power from everything loaded behind it, not the inch. Same here: a box in an avionics bay had to move 1 inch. What made that inch work was a real human catching the problem in the CAD model, 2 years before the first part existed.
2
22
Bruce Lee: be water, my friend. A CAD model still can be. A built jet can't.
5
The failures: hand won't fit behind the box, wrist can't pull 10 connectors, 7 of 10 out of reach. Catch it late and that same inch means a 13-component teardown — including the flight computer — on every jet already built. A 1-hour job becomes a 1-week job.
15
A $3M mockup that never got used once. It was a decommissioned jet, wrong model, and making it usable meant $1M and four months more. We left it parked and ran the cockpit redesign in CAD: 3 (out of 20+) real problems in day 1! Old toys can't learn new tricks.
11
A digital twin doesn't make you see double. It makes you see forty. The flood isn't the failure. Building the ark after the rain starts is. Decide what counts before the session, not after. Then the sim doesn't hand you a flood. It hands you a manifest.
23
The duct was shaped like a W. A tech had to crawl inside it with a scanner. Sim found the scanner loses line-of-sight at the first bend — AND the structure wasn't rated to hold a person's weight. Either bug: ~10 months of delay. We caught both a year before the duct existed.
25
A vendor shipped a door + actuator that installed perfectly and could never come out. Hinge pins pull one direction. Aircraft structure blocked it. Composite airframe, so servicing it means cutting the skin open. $300K to millions. 30 min in sim. Fixed before unit one.
1
15
This is how the lab looked like for this set up
🤖 Made with AI
8
A VR headset gives you eyes and two floating hands. No torso, no spine, no knees. We ran a 5th-percentile female through full-body sim on a reach spec that "passed" a headset walkthrough. Her hands never got close. A headset sees no evil. It still lets you build it.
15