10x faster. That's how much Lightwheel compressed Geely's humanoid training cycle. That's from months to train a task down to weeks. Geely's self-developed humanoid robots are now running on the Auto production line, sorting and sequencing parts alongside human operators. Getting there meant solving the two problems that stall most humanoid deployments: collecting enough real-world data without slowing the line, and closing the sim-to-real gap. Read our latest customer success story on how we pulled this off with Geely! Article: lightwheel.ai/media/geely-hu…

Today, we give robots a /skills library that self-evolves and compounds indefinitely! Introducing ASPIRE: a robot solving its 100th task is no longer as clueless as solving its first. Coding agents observe multimodal sensory traces from simulation and real robots, launch an evolutionary search over control programs, and distill the best know-how into an ever-expanding library. ASPIRE is a new type of continual learning: "training" is skill refinement instead of gradient descent. "Trained model" is a repo of sensorimotor skills instead of floating weights. “Distributed training” is a panel of agents each practicing a different skill instead of sharded minibatches. Here's the beauty: ASPIRE gives the tired terms "sim2real transfer" and "cross-embodiment transfer" a whole new meaning. Bridging the sim-to-real gap is notoriously brutal. An end-to-end policy has to swallow both the visual shift (sim looks toyish next to a real camera) and the subtle contact physics it never quite gets right. ASPIRE sidesteps the mess, because it doesn't ship pixels or weights across the gap, but ships the know-how. The robot still has to practice in the real world, not zero-shot, but it gets there way faster because it isn't rediscovering the strategy from scratch. Same for going single-arm to bimanual hardware, which usually requires new data and retraining from zero. ASPIRE achieves up to ~10x cut in "transfer learning” tokens (yes, tokens are the new unit of *training* compute ;) Check out our gallery of 150 tasks and 90 skills the robots taught themselves, all on the website! Kind of wild that we can ship the "learned weights" as an HTML page rather than a GGUF. We'll open-source the full stack so your own robot library starts compounding from ours! Deep dive in thread:

Wow! I hit 20k subs! Now I need to start posting more often. What content do you want me to lean into most? Tutorials? Talking to experts? Explainers?

Took a chance to introduce A3 to everyone at the @LightwheelAI Physical AI Night during Automate~ @AGIBOT_US

Those hips don’t lie

Who uses business cards anymore? Well, some of you must. These are just from the past month of talking to people IRL. Get out of your office. The facts are in your own office, they are where your customer is.

Huge thanks to @AGIBOT_US for showing up to our @LightwheelAI party at @AutomateShow! The A3 is an incredible robot! I love its fluid motions and stability. @AGIBOTofficial

This would be an awesome apple cannon come Halloween time.

Day 2 at @AutomateShow starts now. If Day 1 was any indication, it's going to be a good one. The floor is open and we're ready at booth 2491. Come talk about what it actually takes to deploy robots in the real world. We're here all day. #AUTOMATESHOW2026 #Robotics

I am pleased to see that ECCV 2026 has desk rejected the papers authored by reviewers who misused LLMs in their reviews of other papers. I applaud the organizers for their courage in following through on this. That said, agentic LLMs have now gotten good enough that the community must now solve the opposite problem: We need to aggressively incorporate AI into the review process. Ideally we should completely rebuild the process around vibe-reviewing with agentic LLMs in some way. If we don't, I expect we will be unable to accommodate the scale and rate of progress in the coming years. The only way out is through.

If you’re headed to @AutomateShow this week and work in Physical AI, don’t miss our Lightwheel party on Wednesday! RSVP, space is limited: luma.com/dzo2klna #automateshow2026

📢 OneCanvas: 3D Scene Understanding via Panoramic Reprojection We extract features from video frames and reproject them into one occlusion-free view of the whole scene that a 2D VLM reads just like a normal image. We can center this view on any viewpoint, including an agent's own pose for situated reasoning. The same projection lets us create spatial training tasks with no human annotation, solvable only by reasoning over the 3D positions of real object features placed on an otherwise empty canvas. The result is a stock 2D VLM that reasons in 3D, setting a new state of the art across spatial benchmarks at far less compute. 🌐 baranowskibrt.github.io/onec… ▶️ youtu.be/NIaHLB9gA7s Great work by @baranowskibrt & @davech2y

This is some excellent 3DGS! 1500 photos is what it takes to get this crazy level of detail from all angles. You could take less, but your scene will degrade from certain viewpoints.

This is the direction physical AI for humanoids needs to go in. Loco-dexterous merges total body control and dexterous movements into one model. That's how humans work. Nice job on Curr-0! @Curr_Robotics 👏

ENPIRE is extremely important to robotics. Every robotics physical AI training pipeline will have an agentic layer in the middle running experiments, evaluating results, and pushing new training loops. This is much faster than human in the loop workflows.

Nobody wants to teleoperate a robot folding laundry 10,000 times. SimWeaver said same. They claim with 200 sim demos on deformable objects, they can hit 91% real success rate on specific tasks. I am keeping an eye on this project. Learn more: simweaver.github.io

I love seeing more open source robots! Especially in the $500 range. This is inexpensive enough that students could use them in their home labs.

I'm headed to the Automate Show and want to connect with the folks working on the frontier of Physical AI. If you are going, leave a comment and lets connect. Also, ask about the Physical AI party I am hosting that Wednesday! #AutomateShow2026 #PhysicalAI #Robotics

I butchered this text :)

We’re seeing the evolution of the UMI. This project made the UME (exoskeleton). They are showing here they he could unsheathe the sword blindfolded using just the tactile sensing in the exoskeleton.