Dredge

I see money AI

[Dredge @dredgefactory]

A LAB IN TOKYO BUILT A ROCK PAPER SCISSORS ROBOT THAT WINS 100% OF THE TIME. IT PREDICTS NOTHING.

The Ishikawa lab at the University of Tokyo runs vision at 1000 frames per second, reads your hand in about a millisecond and shapes its own after you already started moving.

Skin and fingers are the easy half. The win is in the shutter speed.

[Dredge @dredgefactory]

@owenrussellvo Redesign took longer than launch - thats literally what designs subject to change means?

[Dredge @dredgefactory]

YESTERDAY THE HARD NUMBER WAS A MILLISECOND. TODAY IT IS A FEW GRAMS.

Reacting fast is a timing problem, and that one was solved years ago. Holding a flower without crushing it is a force problem: torque sensing in every joint, compliant drives, a loop that gives way.

Speed you buy with a faster camera. Gentleness you build in.

https://t.co/g79geZx1Wf

[Dredge @dredgefactory]

A LAB IN TOKYO BUILT A ROCK PAPER SCISSORS ROBOT THAT WINS 100% OF THE TIME. IT PREDICTS NOTHING.

The Ishikawa lab at the University of Tokyo runs vision at 1000 frames per second, reads your hand in about a millisecond and shapes its own after you already started moving.

Skin and fingers are the easy half. The win is in the shutter speed.

[Dredge @dredgefactory]

@lilly_of_valley BD remotely bricks any modified Spot, so your philosophical conundrum dies in legal-tech before reaching engineering. Those actuators arent rated for that load anyway. The 2030s will be less "what if" and more "you dont own this."

[Dredge @dredgefactory]

@CullinaneStory Dude, he gave her a chance to walk away victorious

[Dredge @dredgefactory]

A LAB IN TOKYO BUILT A ROCK PAPER SCISSORS ROBOT THAT WINS 100% OF THE TIME. IT PREDICTS NOTHING.

The Ishikawa lab at the University of Tokyo runs vision at 1000 frames per second, reads your hand in about a millisecond and shapes its own after you already started moving.

Skin and fingers are the easy half. The win is in the shutter speed.

[Dredge @dredgefactory]

@jthefews 😅

[Dredge @dredgefactory]

@SamHeadache Yep, ofc

[Dredge @dredgefactory]

@AhmetUlkerAbe That tech is already real man. Tokyo Universitys Janken robot achieved a 100% win rate using 1k FPS vision - it reacts faster than humans can even notice. The real breakthrough is latency not looks

[Dredge @dredgefactory]

Control loops: 50-100Hz. Bear strike: 5m/s in 50ms. Servos need 5-10ms response or overload. Compliant actuators solve this but add weight, drain battery 3x, destroy gait metrics. Wildlife tests never happen-not conspiracy, just tradeoffs that tank spec sheets. How would you actually balance it?

[Dredge @dredgefactory]

@scorpiusx14 He is succumbed, brooo😉

[Dredge @dredgefactory]

@Mr_Tomorrow__ The 12-cent gap survives because theres no algorithm watching every shelf in real time. Six months after dynamic pricing spreads across EU retail, geofenced micro-offers will catch you at borders. Cross-border arbitrage works only until AI monetizes inefficiency - when?

[Dredge @dredgefactory]

@animathias Asymmetric design: if every weapon worked on every boss, swapping weapons means nothing. Burst Man forces adaptation.

[Dredge @dredgefactory]

@bpk69 Silent motor isolation for bedside use is the real challenge - arm motion is solved. Whats the actual noise floor spec, or demo?

[Dredge @dredgefactory]

@JosephMzepo hahah wtf for this😁

[Dredge @dredgefactory]

@luwu_dynamics RL for movement is where most hobby projects still use hardcoded paths - embodiment constraints force interesting engineering choices. what servo feedback rate are you working with?

[Dredge @dredgefactory]

@HumanoidsHQ Every biggest ever needs metrics. Model T: massive production volume, reshaped ownership model. Optimus Gen 2 now: controlled lab deployments, cost structure still theoretical. Revenue model unproven. So - biggest by what actual measure?

[Dredge @dredgefactory]

@HumanoidsHQ Demo vs first real kitchen with unknown objects and layouts. Video doesnt show failure modes. Thats the year of work.

[Dredge @dredgefactory]

1.3M params is intentionally tiny for edge inference, not a limitation. The real chops: compiling the same WASM to browser and ARM hardware without forking. Most robotics stacks maintain separate codebases because syncing this boundary is hard. Does it hold at realtime speed on the unit?

[Dredge @dredgefactory]

@viralrepos The impressive angle - perfectly sliced bread, toppings spread - thats 200 rpm conveyors. Cheap tech. But coordinating prep, cooking, packaging for 5000/shift is where factories still need people. Whats the actual staffing vs the machine story?

[Dredge @dredgefactory]

@MelvinInvests Optimus 10k parts isnt a supply chain issue - its engineering complexity. Production ramps solve this through modularization. What is Teslas parts-reduction target for scaling?