Andrea @afelipexyz leads our LATAM efforts — she’s very experienced and incredibly resourceful. We’re also very optimistic about Latin America’s potential, both in terms of data contribution and the strong demand for robotics-related technologies. Andrea lidera nuestras operaciones en LATAM; cuenta con una amplia experiencia y es sumamente resolutiva. También somos muy optimistas respecto al potencial de América Latina, tanto en términos de contribución de datos como de la sólida demanda de tecnologías relacionadas con la robótica.

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 → How it works ↓

Robotics is moving fast. But how close are we to its GPT-1 moment—and what’s still missing? Axis is hosting an X Space, co-hosted with @blocmates and joined by @BitRobotNetwork , @PrismaXai , and @FabricFND , to discuss what it will take to get there. We’ll also share a limited number of access codes with listeners during the Space. 👀 Join us on Friday, July 31 at 11:00 AM PT / 6:00 PM UTC. Set your reminder:

At Axis Robotics, our vision is to build a compounding data engine—one that connects large-scale pretraining data, corrective post-training data, model deployment, and failure feedback in a continuously improving loop. Over the next 6–12 months, we will advance this vision across three connected fronts. On the product side, we plan to scale our egocentric data pipeline in September, with tens of thousands of hours already collected and product requirements being shaped with frontier labs. In October, we will expand our simulation data across more robot embodiments and atomic capabilities. By year-end, we plan to release a large-scale post-training dataset built through human-gated DAgger (HG-DAgger), where the policy acts autonomously and contributors intervene only when it needs correction. On the network side, we will expand our contributor ecosystem into Latin America and Eastern Europe, strengthen our 100K+ contributor network and grow toward 10K DAU. This expansion is designed to support the production of more than 500 hours of egocentric data and 50 hours of simulation data per day while building capacity for corrective post-training data. On the commercialization side, we plan to complete two to three new paid pilots by year-end and work toward becoming a preferred vendor for foundation model companies in Q1 next year. The longer-term goal is to embed the data engine directly into the training and deployment workflows of robot hardware companies, model developers, and industrial operators. These are not separate tracks. They reinforce the same flywheel: broader data coverage produces stronger models; stronger models reach new states; and new failures reveal what data should be collected next. That is the future we are building toward: Scale to produce data continuously. Diversity to reflect the complexity of the physical world. A closed loop to turn deployment feedback and failures into the next round of model improvement. Our north star is not simply more data. It is faster model evolution.

We’re thrilled to announce our investment in @axisrobotics Axis Robotics is cracking the embodied AI data crisis with hybrid data pipelines. Can’t wait to see the “GPT” moment for physical AI.

We couldn’t have achieved any of this without Hack's support from dayone. We’re focused on long-term value and positive externalities. Physical AI is a vast frontier — and we simply want to build something genuinely useful in it

Congratulations to @axisrobotics on their $12M Seed round. Axis is building one of the leading scalable physical AI & robotics platforms on @base, pioneering the Train-to-Earn era. They’re also: 🟦 Finalists in the @base Batches 003 Virtuals Robotics Track 🟦 Part of Base Founders Residency Batch 002 Excited to see what’s next for the team. Check out what they’re building 👇

Grateful to all our investors for the trust and support. Your backing is the strongest validation of our team and the community building with us. Physical AI is a massive opportunity, but we’re still early. Our goal is simple: move fast, stay grounded, and create real value at every step. Onward!

🚨 AMA Announcement 🚨 Today at 9PM KST / 8PM SGT 🎙 MC: Humbleman (@Cryptowombat125) 🎤 Speaker: Chris (@chris_anm01), CEO of @axisrobotics Join us to hear the vision behind Robotics General Intelligence (RGI) and the future of Physical AI. 📍 X Space: 🕘 9PM KST / 8PM SGT today Don't miss it!

The dataset and paper based on the previous phase of community contributions are now live： Paper Link: Project Page: Dataset Link: Github Codebase: Using Pi0.5 + AXIS-100%, we achieved 88.8 overall success on LIBERO-Plus. For comparison: Vanilla Pi0.5 achieved 83.9 A RoboCasa-matched simulation baseline achieved 57.5 This clearly demonstrates that diverse, in-the-wild data provides substantial gains in model performance. And V2 is already in progress — significantly larger in scale, covering more embodiments and a wider range of atomic capabilities.

What V1 proved: An end-to-end sim-to-real pipeline works. Engineered diversity beats raw volume (79.4 on LIBERO-Plus), and transfers directly to real physical robots. The bottleneck: V1 is unidirectional. Once deployed, the model can't recover from its own failures — because the data flow stops after training. The V2 shift: Close the loop. Human-Gated DAgger turns model errors into targeted data collection. Axis is no longer a pipeline — it's a compounding, human-in-the-loop data engine.

We're building a Task Generation System for robot learning. Four design principles from day one: Structured — Tasks are defined through clear data hierarchies and parameterization, not loose script stacking. Productized — This isn't an internal tool. It's a configurable, deliverable product that non-engineers can understand and operate. Extensible — New scenarios, new capabilities, new asset classes — all additive, never requiring a rebuild. API-first — The end goal is open access. Interface design, parameter specs, and output formats are built for external consumption from the start. This is the engine behind data diversity