$BASLTION is live on @RobinhoodCrypto.
CA: 0x79AA443BF74a1b39E3FC08115FB4aA8F19E7c9a6
What the site actually runs: a Unitree G1 humanoid in MuJoCo, compiled to WebAssembly and running on your machine, not streamed from a server. Open the rig and move all twenty-nine joints by hand, or load the current policy and drive it with the joystick.
Every training generation is published as a file: the policy, the rollout frames, the metrics, the sha256. The replay you watch in the browser is the exact rollout that was recorded on the training box, and you can hash it yourself to prove it.
Training started today at generation 0. It has no trained policy and it falls in about a second. A new generation lands every few minutes, and nothing reaches the page until it survives a hundred evaluation episodes on fixed seeds.
https://t.co/VUHsSDTter
The loop is closed.
Before every training segment the trainer reads the seats contract on chain and applies whatever the holders set. Burn into a joint, take it, turn it down, and the next segment trains with your number.
An unclaimed joint runs at one hundred percent. A held one runs at whatever its owner decided.
https://t.co/H6BGk9zWv6
Basltion is open source.
https://t.co/Zyqz1bB668
The trainer, the browser simulator, both contracts and every test are in there. Clone it and run the robot on your own machine.
The seats contract has no owner, no pause switch and no upgrade path. You do not have to take our word for it. The source is right there, MIT licensed.
https://t.co/VUHsSDU13Z
Less than two hours ago it could not stand up.
It stood at generation 3. It walked at generation 5, fourteen metres without going down.
Eighty generations later it is still failing at rough ground, seventy percent where it needs ninety, and every failed attempt is published as it happens.
https://t.co/q7HbiAlftW
The robot trains while you sleep. It does not take weekends and it does not care what the chart is doing.
It falls, gets up, trains again, and publishes whatever it learned. Every few minutes, around the clock, whether anyone is watching or not.
https://t.co/34Vnw88RvQ
The seats are open.
The contract is live on Robinhood Chain. Twenty-nine joints on one body, one owner each.
Burn more $BASLTION into a joint than the current holder burned and the seat is yours. The burn is gone the moment it lands, whether you hold the joint for a month or lose it in an hour. The contract states that twice, in writing.
Holding a joint gives you one lever: the torque budget behind it, anywhere from seventy to one hundred percent. Set it and the next generation trains with what you chose. Turn a knee down and the robot has to learn to walk with a weaker knee, in public, in front of everyone.
Seats: https://t.co/2oEwPfQKT5
https://t.co/H6BGk9AukE
Every generation the robot trains gets published as a file, and the owner of each joint is written inside it. The file gets hashed, and the hash goes on chain.
So your address ends up inside the record of a machine learning to walk. Not a leaderboard that wipes every season. A file that nobody can edit afterwards, including us.
https://t.co/VUHsSDTter
$BASLTION is live on @RobinhoodCrypto.
CA: 0x79AA443BF74a1b39E3FC08115FB4aA8F19E7c9a6
What the site actually runs: a Unitree G1 humanoid in MuJoCo, compiled to WebAssembly and running on your machine, not streamed from a server. Open the rig and move all twenty-nine joints by hand, or load the current policy and drive it with the joystick.
Every training generation is published as a file: the policy, the rollout frames, the metrics, the sha256. The replay you watch in the browser is the exact rollout that was recorded on the training box, and you can hash it yourself to prove it.
Training started today at generation 0. It has no trained policy and it falls in about a second. A new generation lands every few minutes, and nothing reaches the page until it survives a hundred evaluation episodes on fixed seeds.
https://t.co/VUHsSDTter
Token launch is coming to @RobinhoodCrypto
The robot has been learning in the open since the first day. 102 generations published, each one with its hash, its survival rate and the distance it covered. Nothing backfilled, nothing cherry picked.
https://t.co/34Vnw88RvQ
Basltion is live on web.
A Unitree G1 humanoid is learning to walk in MuJoCo, in public. Every training generation gets published with its hash, its survival rate and how far it actually got. Generation 0 fell in under a second. Generation 102 walks.
Open the rig in your browser and move all 29 joints yourself. Replay any published generation. Shove the robot mid-stride and see whether the policy catches it.
Nothing is pre-rendered. The physics runs on your machine.
https://t.co/VUHsSDTter
Most robot demos show you the take that worked.
We published every generation, failures included. Standing came at generation 6, walking at 7, rough ground took another 84.
Here is how the whole thing runs. https://t.co/L2CZ1cdhRk
Training started today.
Basltion is a Unitree G1 humanoid learning to walk in MuJoCo physics. Every generation is published with the sha256 of its files, so anyone can check what it did.
Generation 1 fell in under a second. Generation 6 stood through twenty seconds in 98 of 100 episodes. Generation 7 walked 13.5 metres with a velocity tracking error of 0.15 m/s.
It is now failing on rough ground. That is the next stage.
Introducing NVIDIA Jetson Orin Nano 2 👏
Built to bring frontier intelligence to entry-level edge AI, this robotics computer will deliver:
• 2x the inference performance of Jetson Orin Nano Super
• 40% lower power consumption at the same performance in 15W mode
• The same compact form factor
Read the press release 👉 https://t.co/d3nGTlFPBD
🚨 THE STRONGEST ROBOT GRIPPER ISN'T HYDRAULIC - IT'S POWERED BY TWISTED STRING 🤯
This is a Twisted String Actuator. Two strings twist together, contract, and slam those jaws shut with insane force.
Top is the CAD, bottom is the real thing - smaller than your finger but can hold HEAVY weight. No motors in the fingers, no hydraulics, just string. This is how next-gen robotic hands and prosthetics will lift 100x their weight.