Pico | (✱,✱)

Based Contributor | Defi Explorer. 🎒

Contributing in the little ways I can. 🧡

[Pico | (✱,✱) @_Picokel]

I went researching on DR just like @RWenwenwen.

I realized my definition wasn't specific, and my image was wrong. ✖️❌

Yes, Domain Randomization(DR) typically involves manipulating physical conditions of the environment and objects during replay.

But it's not that simple. 🙃

Here's how it works:

1. Teleoperation + Data Storage:

After teleoperating the robots to complete a task, data on the robot's joint positions, end-effector(gripper) states,.. etc during tasks are stored in an Axis Unified Trajectory format.📦

Saving it in this format allows it to be easily replayed within Axis AI while scaling the training data.

2. Replay & Domain Randomization:

This saved trajectory data contains the robot's joint coordinates, gripper actions,. etc. at specific times and has to be replayed in a more photorealistic environment for training. 🏕️🏞️🎥

While replaying trajectories (regenerating demonstrations) in a more photorealistic environment (NVIDIA Isaac Sim), alterations are made to the demonstrations, in:

• Object mass, textures, slipperiness, friction.✳️
• Environment lighting. 🔅

(These changes in physics, objects & environmental features is Domain Randomization).

Due to domain randomization + the replay (cross-simulator replay), the replayed demonstrations are more varied and photorealistic, such that problems like:

"A robot unfamiliar with lifting heavy objects/materials drops them due to insufficient force generation." are avoided, because it is familiar with heavy mass already...

So, these more photorealistic demonstrations are what are presented to the Vision-Language-Action (VLA) models to train them.🦾

[Pico | (✱,✱) @_Picokel]

𝗗𝗼𝗺𝗮𝗶𝗻 𝗿𝗮𝗻𝗱𝗼𝗺𝗶𝘇𝗮𝘁𝗶𝗼𝗻: altering of conditions in collected manipulation trajectories to diversify it.

[Pico | (✱,✱) @_Picokel]

Introducing Fluton.

- Fluton is a cross-chain execution layer that makes all onchain interactions private by default. 🗽

- It obscures transactional data, helping to protect user onchain activity, balances and strategies at every part of an onchain interaction. 👀

- Fluton achieves this by use of FHE (Fully Homomorphic Encryption). 🫆

- FHE enables this: encrypt transactions, and process on encrypted data without a need for decrypting.

Fluton's privacy infrastructure is beneficial to financial institutions seeking private execution onchain, onchain users seeking private payments, large investors and more. 🏛️⭕

See → @FlutonIO

[Pico | (✱,✱) @_Picokel]

More memes, fam. 😅

See more 👇🏼

[Pico | (✱,✱) @_Picokel]

Axis robotics NG meme submissions. 😅↓

See more below ↓

@iamlogtun

[Axis Robotics @axisrobotics]

GAxis (✱,✱)☀️

New week. New challenges. New access.

We’re giving away 10 BitRobot access codes over the next 72 hours.

Winners will get access to SN/04 and start earning rewards from both Axis and BitRobot.

To join:

1. Follow @axisrobotics & @BitRobotNetwork
2. Like + repost this post
3. Comment with a screenshot or photo of where you’re training right now

Grinding tasks? Climbing the leaderboard?

Show us your journey.

[Pico | (✱,✱) @_Picokel]

First week of June. 🍃

PS: This is Community effort, not official.

See @axisrobotics for official updates.

[Pico | (✱,✱) @_Picokel]

Happy New Month. ❤

For the 24th - 30th May, 2026 week. 📆

[Pico | (✱,✱) @_Picokel]

@axisrobotics How did it go?

[Pico | (✱,✱) @_Picokel]

@PMemoye @choc4lat @Jez_Cryptoz The fire should be kept lit. 🕯️

[Pico | (✱,✱) @_Picokel]

B U R N

We walked 'round the fire, time after time. ⏳

Thinking we would get burnt. 🔥

But Instead, we were formed 🪨

Through the Agneepath, we walked; and got ritualized. 🐦‍🔥

gRitual. 🕯️

Have a great week, Community 🍃

[Pico | (✱,✱) @_Picokel]

@YoungestGrandad @arc @jerallaire Man, you're shipping on Arc too?

[Pico | (✱,✱) @_Picokel]

@DefiIntro Is the Zentra official account back yet?

[Pico | (✱,✱) @_Picokel]

Axis robotics NG meme submissions. 😅↓

See more below ↓

@iamlogtun

[Pico | (✱,✱) @_Picokel]

@Xinam555 Damn bro, details

Kudos for your contributions though bro

[Pico | (✱,✱) @_Picokel]

@Aemar00 @iamlogtun @axisrobotics @Larrizle_no1 Congratulations

[Pico | (✱,✱) @_Picokel]

@Larrizle_no1 @iamlogtun @axisrobotics @Aemar00 Congrats bro 👏🏼

[Pico | (✱,✱) @_Picokel]

@Siggyworld Okay, I'm here

Hello Siggy

[Pico | (✱,✱) @_Picokel]

@akinmarve26 How do I get started on this, bro?

[Pico | (✱,✱) @_Picokel]

@wiimee @amit0xic Huddle01 is a suspicious app, man?

[Pico | (✱,✱) @_Picokel]

@akinmarve26 That's a good development, buddy

[Pico | (✱,✱) @_Picokel]

@BigNelson_ @axisrobotics @0xzagen @MPriosin71748 Yes!!

[Pico | (✱,✱) @_Picokel]

Cross-simulator replay on @axisrobotics. 🗺️

The axis robotics web-based simulations use MuJoCo, a lightweight simulator engine. 🕸️

The MuJoCo simulation used is appreciated for ease of access and running through a browser, but there exists a flaw:

'It doesn't accurately reflect real-world photorealistic environments good for VLA model training...' 🗺️

So, robotic trajectories generated on MuJoCo have to be replayed on a more photorealistic simulator engine (likely NVIDIA Isaac Sim), where objects & environment would appear more photorealistic just like they are in real life. 🏕️

Why?
VLA models which process, learn and perform from what they see cannot generalize on unrealistic data; they could get stuck operating in the real world.

However, native MuJoCo trajectories do not just replay in Isaac Sim.🙎🏼

So, Axis employs its Unified Abstraction Layer, and Unified Trajectory format.

Following this layer, Trajectory data is stored in a unified trajectory format. 💾

Using this format, the downstream replay simulator can replay stored trajectory data without getting confused by the different naming language/conventions of the simulator where the trajectory data was generated at. 🙂

This data storage format makes it possible for:

- The NVIDIA Isaac Simulator to replay MuJoCo trajectories (not perfectly though).🎬

- Downstream VLA models to understand the data during training. 👍🏼

During this cross-simulator replay, domain randomization (see quoted tweet) is applied to scale up data quality.

However, cross-simulator replay doesn't promise to reproduce perfectly same execution from MuJoCo.

It promises replay of the same seQueNce of events in completing task from MuJoCo; in a more photorealistic scene (due to DR).

Please, confirm from https://t.co/fBNukx4RyN if any doubts.

[Pico | (✱,✱) @_Picokel]

I went researching on DR just like @RWenwenwen.

I realized my definition wasn't specific, and my image was wrong. ✖️❌

Yes, Domain Randomization(DR) typically involves manipulating physical conditions of the environment and objects during replay.

But it's not that simple. 🙃

Here's how it works:

1. Teleoperation + Data Storage:

After teleoperating the robots to complete a task, data on the robot's joint positions, end-effector(gripper) states,.. etc during tasks are stored in an Axis Unified Trajectory format.📦

Saving it in this format allows it to be easily replayed within Axis AI while scaling the training data.

2. Replay & Domain Randomization:

This saved trajectory data contains the robot's joint coordinates, gripper actions,. etc. at specific times and has to be replayed in a more photorealistic environment for training. 🏕️🏞️🎥

While replaying trajectories (regenerating demonstrations) in a more photorealistic environment (NVIDIA Isaac Sim), alterations are made to the demonstrations, in:

• Object mass, textures, slipperiness, friction.✳️
• Environment lighting. 🔅

(These changes in physics, objects & environmental features is Domain Randomization).

Due to domain randomization + the replay (cross-simulator replay), the replayed demonstrations are more varied and photorealistic, such that problems like:

"A robot unfamiliar with lifting heavy objects/materials drops them due to insufficient force generation." are avoided, because it is familiar with heavy mass already...

So, these more photorealistic demonstrations are what are presented to the Vision-Language-Action (VLA) models to train them.🦾

[Pico | (✱,✱) @_Picokel]

@get_optimum Welcome to Optimum, Lewej