We’ve raised $50 million in Series A funding from DST Global Partners, @nvidia (NVIDIA’s venture capital arm), @redalpine , @prosusventures , and @Moonfire_VC , following our $7.35M seed round from @frst_vc , @Moonfire_VC , and @redalpine just a few months earlier, to build the

I want to give a huge thanks to @un1c0rnioz for kicking off the project and driving it forward. To @asoare159 for the mentorship and the late-night debugging sessions. To @ihorbeaver for supporting the initiative from day one. And of course to @AractingiMichel and the

Here’s a great visualization of how RTC works. You can see two versions side-by-side — with and without RTC. Each line represents one flow-matching step, and you can clearly see how RTC guides the early actions of the new chunk to align smoothly with the previously generated

Great news 🚀 It took a while, but we finally made it! 🤗 Real-Time Chunking is finally merged into @LeRobotHF ! PR: https://t.co/Vsop7JN6Ws The RTC method comes from the amazing team at Physical Intelligence: Paper: https://t.co/SL9fVvPyiL Now the implementation is available

We just released results for our newest VLA from Physical Intelligence: π*0.6. This one is trained with RL, and it makes it quite a bit better: often doubles throughput, enables real-world tasks like folding real laundry and making espresso drinks at the office.

Introducing VLAb: a pretraining toolkit for Vision-Language-Action models for robotics! Super simple and hackable codebase to experiment with VLAs! Take any LeRobot/HF format dataset and pretrain or fine-tune VLA models efficiently across many GPUs. 🧵 Here’s what it includes:

If you’re building robotics models, take a moment to check this discussion: https://t.co/WM3iYLrduS The @LeRobotHF team is collecting feature requests and feedback for Datasets. Data collection and tooling are still far from perfect today — so this is a great opportunity to

What the great time we live - building space datacenter in cosmos is a super close reality 🤠

1X announcing home robots available for delivery in 2026 😮 This is incredible — a huge milestone not just for 1X, but for the entire robotics industry. It’s going to kick off an exciting wave of announcements from Figure, Tesla, and others soon. I’m really curious about the

I am starting a venture on top of LeRobot! We’re at a pivotal time. AI is moving beyond digital to the physical world. Embodied AI will change our surroundings in ways we can barely imagine. This technology holds the potential to empower everyone. It must not be controlled by

It's the most desired feature everyone has wanted for months and tried to implement themselves. 😁

Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single,

Thats sick! It’s how our world Should look like.

The pi0.5 is merged into LeRobot's main branch. https://t.co/0QjzKRtaJW Great shoutout to @pepijn2233 and @LeRobotHF team! 🔥

🚀 The @GoogleDeepMind just announced a new wave of VLA-related models: 🔗 https://t.co/4iqP6yxIga Papers: Tech Report: - https://t.co/8uAHPAibeq - arXiv: https://t.co/pUvIAKZb9k What I like: - VLMs trained directly on robotics data & tasks → exactly where the industry is

🏆 LIBERO VLA Leaderboard is live — the first-ever leaderboard for Vision-Language-Action models! VLAs are growing fast, so we need efficient, fair evaluation on shared benchmarks and an open community space. Now it’s finally possible 👇

✨ New in LeRobot ✨ We now officially support LIBERO, one of the largest open benchmark for Vision-Language-Action (VLA) policies with 130+ tasks 🤯 Why this matters: 🧩 Unified benchmark: evaluate any VLA policy under a common setup. 🛠️ Easy integration: just install

The next step with Hil-SERL - vla RL fine tuning. @_lilkm_ extended the pipeline to follow ConRft paper 🤘🎊 Great job.

pick-and-place cubes on a SO100 arm using the @LeRobotHF with a QC-FQL policy I am currently implementing. It is based on the HIL-SERL implementation in LeRobot ...

🚀 Big news! The new VLA is now open-weighted. The pi0.5 release marks a significant shift — moving from training the Action Expert for solving robotics tasks toward the VLM component. This version is expected to be smarter than the previous pi0, and importantly, fine-tuning