Axis Robotics

Scale Physical AI for the real world.
Robot intelligence is not built by a few; it's built by all.

[Axis Robotics @axisrobotics]

Axis is officially LIVE on @base. 🔵

Axis is scaling Physical AI for the real world, contributed by everyone.

You can control robots in a virtual world, generate training data at scale, and help build the brain behind tomorrow's robots. All from browser. No hardware needed.

Start building robotics intelligence today: https://t.co/xeQnZzV2WM

[Axis Robotics @axisrobotics]

@Nibel_eth @base

[Axis Robotics @axisrobotics]

Community AMA in two minutes

https://t.co/dKOfdQgfi5

Come and ask the team anything you want to know!

[Axis Robotics @axisrobotics]

Axis Weekly

Last week, we focused on moving recently completed infrastructure into reliable production use.

We deployed established benchmark tasks, accelerated verification, improved cross-embodiment consistency, and made TaskGen (Task Generation) and HG-DAgger (Human-Gated DAgger) more reliable for larger post-training runs.

Key updates:

- Benchmark deployment: 50 LIBERO Pro tasks are now online, and the first four RoboCasa tasks were validated.

- Verification: Caching and worker upgrades roughly tripled verification speed while improving scoring and false-positive handling.

- Embodiment alignment: OpenArm issues were resolved, T1 sim-to-real behavior was aligned, and the Axis-to-RoboVerse task-conversion path was completed.

- TaskGen and HG-DAgger: Randomization expanded, correction data became more consistent, and collection and evaluation throughput improved.

Details below ⬇️

[Axis Robotics @axisrobotics]

Axis Weekly

This week focused on getting DAgger ready for launch, expanding TaskGen’s capabilities, and bringing RoboCasa scenes into the browser.

We added batch post-task generation, tightened the operation flow, introduced new replay-verification tools, integrated OpenArm into TaskGen, and began automating asset generation.

Key updates:

- DAgger launch preparation: Batch post-task generation was added, alongside operation-page fixes that close interface loopholes during correction-data collection.

- Replay verification: DAgger now supports seed-based object reset and verification, with replay validation being applied to collected data.

- TaskGen upgrades: Generation performance improved, auto-checker coverage expanded, and a wider range of initial-position randomization was enabled.

- OpenArm and AssetGen: OpenArm was integrated into TaskGen, and work began on an automated AssetGen worker and a more compact asset taxonomy.

- RoboCasa in the browser: RoboCasa scenes were migrated into the web task interface, with previous lag issues addressed.

Details below ⬇️

[Axis Robotics @axisrobotics]

TaskGen Randomization and HG-DAgger Reliability

TaskGen expanded both spatial and variant randomization. Spatial sampling now covers a wider and more uniform distribution instead of a small number of tightly clustered points.

New visualization and validation tools show whether generated initial states cover the intended workspace or collapse into narrow modes.

For HG-DAgger, frontend observation-history construction and action-application semantics were aligned. This makes collected corrections more consistent with policy execution.

Drag-control frequency was reduced to make command timing more consistent. Caching, process reuse, and batched inference also increased collection and evaluation throughput for larger post-training runs.

[Axis Robotics @axisrobotics]

@base Yes—even robot data and models.

[郡主Christine (✱,✱) @0xsexybanana]

Live on @base for four month now, and we have collected over 2M data trajectories and more than 100,000 contributors are building the future of Physcial AI with us. Base is the cradle for Physical AI.

@axisrobotics reduces the entry barriers of robotic data acquisition by enabling web-browser-native access, and crowdsourced simulation datasets have proven viable for robotic learning.

Only possible on @base

[Axis Robotics @axisrobotics]

Thanks @baseapac for having us.

Physical AI is hitting a data wall. On @base, we’re building the infrastructure to scale beyond it.

Axis brings together browser-based simulation, real-world egocentric capture, and a global contributor network to produce diverse training data at scale.

Watch our co-founder & CMO @0xsexybanana unpack why we started Axis, what we’ve built so far, and what’s next. ⬇️

[Base APAC @baseapac]

Base Stories Episode 005: Axis Robotics

@0xsexybanana started her career creating ice cream flavours at Häagen-Dazs.

That experience taught her design thinking and sparked her passion for startups.

Today, she's building one of the leading physical AI companies on @base with @axisrobotics.

Watch till the end to hear her founder journey, why building through a bear market became an advantage, and where @axisrobotics is headed next after closing its recent $12M Seed round. 👇

[Axis Robotics @axisrobotics]

@baseapac @0xsexybanana @base based

[Axis Robotics @axisrobotics]

Read the complete Points rules and scoring mechanism:
https://t.co/fcwC8LD9WP

[Axis Robotics @axisrobotics]

The Axis Point System is LIVE.

Every valid data contribution is recorded, quantified, and reflected in your Points—so real contributors are recognized for honest, high-quality work that advances robotics.

Check your Points → https://t.co/o543a0DNbW

How it works ↓

[Axis Robotics @axisrobotics]

Points for Early Axis Contributors

Valid contributions signed before 15:59:59 UTC on July 28, 2026, received a one-time retroactive Points grant—nothing to redo, nothing lost. Your Discord roles and earned badges will not be converted into Points.

Submissions signed after that snapshot will count toward the Epoch in which they are signed.

[PL 漂亮 (✱,✱) @plpiaoliang]

Community AMA in @axisrobotics discord

Date: 4th Aug 2026 10PM SGT

@chris_anm01 @0xsexybanana will be joining us too! 👀

[Axis Robotics @axisrobotics]

@Rnoozy (✱,✱) mode on