Azka Ramdani

Contributor @AxisRobotics

[Azka Ramdani @ARamdani150]

A robot dataset isn’t just a collection of trajectories.

It’s a collection of tasks.

Different tasks create different motion patterns, skill compositions, and behaviors.

That task structure is where much of the learning value comes from.

[Axis Robotics @axisrobotics]

https://t.co/v8VgqcSfOn

[Azka Ramdani @ARamdani150]

Axis doesn’t sell the same thing to every customer.

For model companies, it may be a task-grounded training dataset.

For hardware companies, it may be a full data → model → deployment workflow.

Different deliverables.
Same data engine.

[Axis Robotics @axisrobotics]

https://t.co/v8VgqcSfOn

[Steven Ferdinand (✱,✱) @Stevennn77_]

NIh aku sharing buat yang penasaran gimana sih cara kontribusi atau grinding di project web3 yang fokus ke robotika atau Physical AI ✨👌

Disini aku kasi liat visual saat ngelakuin teleoperasi pake 3 project yang lagi aku serius jalanin :

🤖 @axisrobotics
🤖 @BitRobotNetwork
🤖 @ZenO4AI

Pada masing-masing project ini kita nnti akan diberi tugas yang harus diselesaikan, contohnya kyk taruh wortel dipiring, nyusun balok supaya sejajar, angkat balok dan sebagainya dari yang gampang smpe sulit 😭

Intinya kita cuma perlu nyelesaiin tugas sesuai perintahnya aja, awalnya pasti sulit aku juga ngerasain gitu 😭 tapi kalo kalian udh paham mekanismenya pasti bisa kok 👌

Tapi yang perlu kalian ketahui masing-masing project punya mekanisme dan kontrol yang berbeda jadi kalian harus bisa beradaptasi 😉

Semangat grindingnya ya kawan ✨🔥

[Azka Ramdani @ARamdani150]

@Stevennn77_ @axisrobotics @BitRobotNetwork @ZenO4AI mantap bang tutornya

[Azka Ramdani @ARamdani150]

Why does layout randomization matter?

Because robots shouldn’t memorize where objects are.

They should understand the task regardless of scene arrangement.

@axisrobotics is showing measurable gains from that approach.

[Azka Ramdani @ARamdani150]

The biggest AXIS gain wasn’t from a bigger model.

It came from better data diversity.

A +23.2 point improvement on Layout robustness suggests that structured scene variation can be just as important as model scaling.

[Axis Robotics @axisrobotics]

The largest gain comes from Layout, the axis where structured scene diversity matters most.

LIBERO-Plus evaluates robustness across seven perturbation axes: Camera, Light, Sensor Noise, Background, Layout, Language, and Robot.

The largest absolute gain from AXIS-100% appears on Layout: +23.2 points over vanilla π0.5.

This directly validates one of the core AXIS data augmentation methods of layout randomization.

AXIS also improves more than visual robustness.

Beyond Layout, AXIS-100% also improves several other LIBERO-Plus axes. These axes measure different kinds of distribution shift.

The result shows that AXIS also improves robustness to viewpoint shifts, degraded visual observations, robot initial-state changes, and task wording variation.

[Azka Ramdani @ARamdani150]

@Rafiadehanafi @axisrobotics Menanti next event

[Azka Ramdani @ARamdani150]

@Hamza_Abi16 gAxis

[Azka Ramdani @ARamdani150]

@airdroppinter @AxisRoboticsID LFG!!

[zerokids @zerokidsnft]

Shoutout to all the amazing projects and communities we've collaborated with so far!

Take a look at our growing collab wall: https://t.co/SinJOGNSJf

Thank you for being part of the ZeroKids journey.

[Azka Ramdani @ARamdani150]

A useful reminder:

Robot intelligence isn't only about model size.

How you generate and randomize data matters.

AXIS-100% delivered its largest gain on Layout robustness, highlighting the value of structured scene diversity.

[Axis Robotics @axisrobotics]

The largest gain comes from Layout, the axis where structured scene diversity matters most.

LIBERO-Plus evaluates robustness across seven perturbation axes: Camera, Light, Sensor Noise, Background, Layout, Language, and Robot.

The largest absolute gain from AXIS-100% appears on Layout: +23.2 points over vanilla π0.5.

This directly validates one of the core AXIS data augmentation methods of layout randomization.

AXIS also improves more than visual robustness.

Beyond Layout, AXIS-100% also improves several other LIBERO-Plus axes. These axes measure different kinds of distribution shift.

The result shows that AXIS also improves robustness to viewpoint shifts, degraded visual observations, robot initial-state changes, and task wording variation.

[Azka Ramdani @ARamdani150]

One of the strongest AXIS results:

The biggest improvement came from Layout robustness.

On LIBERO-Plus, AXIS-100% achieved a +23.2 point gain over vanilla π0.5 on the Layout axis.

More scene diversity.
Better generalization.

[Axis Robotics @axisrobotics]

The largest gain comes from Layout, the axis where structured scene diversity matters most.

LIBERO-Plus evaluates robustness across seven perturbation axes: Camera, Light, Sensor Noise, Background, Layout, Language, and Robot.

The largest absolute gain from AXIS-100% appears on Layout: +23.2 points over vanilla π0.5.

This directly validates one of the core AXIS data augmentation methods of layout randomization.

AXIS also improves more than visual robustness.

Beyond Layout, AXIS-100% also improves several other LIBERO-Plus axes. These axes measure different kinds of distribution shift.

The result shows that AXIS also improves robustness to viewpoint shifts, degraded visual observations, robot initial-state changes, and task wording variation.

[EnpiciFi (✱,✱) @masssnpc]

jangan lupa yakk besok ramaikan warga @axisrobotics @AxisRoboticsID , dah intinya Pastikan akun GeoGuessr kalian sudah siap buat besok.........

[Azka Ramdani @ARamdani150]

@masssnpc @axisrobotics @AxisRoboticsID Siap bang!

[Azka Ramdani @ARamdani150]

Physical AI is powered by people.

Axis and BitRobot are opening access to SN/04, giving contributors a chance to generate robotics training data and earn rewards across both ecosystems.

Train. Contribute. Climb the leaderboard. 🤖
@axisrobotics

[Azka Ramdani @ARamdani150]

Axis is turning robotics training into a repeatable service.

Task ID → train model → evaluate → generate heatmaps → deploy.

Less manual work.
More scalable experimentation.
@axisrobotics

[Azka Ramdani @ARamdani150]

@masssnpc @axisrobotics gAxis

[Azka Ramdani @ARamdani150]

@0xRahvanaa gass bang

[Azka Ramdani @ARamdani150]

@HikalHank @axisrobotics gAxis bang

[Azka Ramdani @ARamdani150]

One interesting result from Axis:

Recovery-driven training improved performance by ~10% over baseline.

The next question:
How much comes from more data vs better data?
That’s what the new failure-task experiments aim to answer.

[Axis Robotics @axisrobotics]

Axis Weekly

This week was about making the AXIS loop more scalable end to end: automating data-to-model workflows, testing recovery-driven training, expanding TaskGen coverage, and preparing the dataset and model stack for release.

Key updates:

- Data-to-model automation: We used scripts to speed up and standardize several repetitive but critical workflows.

- Continuous-growth training: We completed multi-data-scale training and success-rate comparisons across several failure tasks.

- Failure task expansion: A new batch of failure tasks has been pushed to test, expanding the evaluation range for ablations across data scale, data quality, and randomization.

- TaskGen: Articulated-object generation is now merged into the automatic generation pipeline.

- Model and release prep: We finished the first round of fine-tuning, evaluation, and benchmarking, completed the dataset’s conference submission, and are now improving experimental results for release.

Details below 🧵