Good morning Legends 🏜️ Have a wonderful day 💞 Do you need alpha..? If need check it👇 @Canborsa_DEX OG Program is now live! 👀 Early supporters and active community members can unlock some exciting perks ... - 2.5× Canborsa Points boost . - 30% referral boost . - Telegram Premium gifts . - Private OG chat . - Exclusive creator campaigns . - Direct access to the team . Full details 👇 t.me/canborsa_dex/185

What if the biggest problem in robotics isn’t a lack of data, but how robots learn from it? Building a capable robot takes more than collecting thousands of demonstrations. The real challenge is turning that data into skills that work beyond controlled environments. Axis Robotics is exploring this problem from four important angles. < 1. Imperfect Data Can Still Teach > Not every human demonstration is perfect, and not every trajectory is equally useful. But diverse data can expose models to different behaviors and ways of completing a task. The goal is to learn useful patterns, not simply collect flawless examples. < 2. Simulation + Real-World Experience > Real-world demonstrations can be expensive to collect. Simulation offers another way to expand the training data. In an Axis Robotics study, selected simulation data helped raise average real-robot success from 16.7% to 85.8% across three tested manipulation tasks, using 10 real-world demonstrations per task. Interestingly, using the full simulation dataset reached only 25.8%. The lesson? More data isn’t automatically better. The right data matters. < 3. Success Is Not the Same as Reliability > A robot completing a task once doesn’t mean it has truly mastered it. Different object positions, unfamiliar conditions, and unexpected mistakes can expose weaknesses that weren’t obvious during training. That’s why evaluation matters. Robots need to be tested beyond the conditions where they already succeed, so researchers can identify what still needs improvement. < 4. Every New Task Should Build on Previous Learning > What if a robot didn’t have to start from scratch every time it faced a new task? Axis Robotics work with Booster Robotics explores how simulation, real demonstrations and continued pretraining can create a stronger starting point for future tasks. Instead of treating every task as an isolated experiment, useful knowledge can become a foundation for what comes next. < THE BIGGER PICTURE > These four ideas point toward one direction: → Collect diverse experiences → Identify the data that matters → Test and learn from failures → Reuse knowledge across future tasks Physical AI isn’t just about building bigger datasets. It’s about creating a learning process where better data leads to better training, real-world testing reveals what’s missing, and every new task can benefit from what came before. That’s the part of Axis Robotics I find most interesting: building a data engine that can help robot learning improve over time. @axisrobotics #AxisRobotics #PhysicalAI #RobotLearning

Not every successful robot demonstration is worth learning from. The real value lies in the correction that actually makes a difference. Axis Robotics is exploring an interesting idea: teaching robots through short, verified human corrections instead of blindly copying every action. < WHAT MAKES A CORRECTION USEFUL? > - The robot genuinely struggles with a task - A human steps in to correct the mistake - The key action is isolated and tested - The robot can continue and finish the task on its own < WHAT DID AXIS FIND? > According to Axis’s published research: - 660 correction attempts were collected - Only 161 short correction snippets passed the full verification process - Each selected snippet was around 0.8 seconds long Why does this matter? Because a long demonstration can include unnecessary movements, pauses, and actions that don’t actually help the robot learn. More data doesn’t always mean better learning. Sometimes, a small amount of carefully verified data can be far more useful than a much larger collection of raw demonstrations. < THE BIGGER PICTURE > Axis is working toward a system where community contributions help identify a robot’s weaknesses and teach it how to recover. The goal isn’t just to collect more data. It’s to make every useful correction count. That’s what makes this approach worth watching as Physical AI continues to evolve. @axisrobotics @KaitoAI #AxisRobotics #PhysicalAI

Something interesting is taking shape at @perpsxtrade 👀 PerpsX is building a perpetuals trading terminal powered by @Polymarket Perps. >> Bringing crypto equities, indices. >> Commodities into one place, with features like up to 20x leverage. >> Flexible margin options. But it's not just about trading. The team is also looking toward community growth with potential opportunities in ambassador, community management, and other roles for people who actively contribute. Join the community: t.me/perpsx Website: perpsxdex.xyz/ And here's the latest update: PerpsX plans to open 6 paid mod positions, with 2 Telegram mods and 4 Discord mods. 👀 Don't miss it, Check comment..

A robot doesn’t necessarily need one giant model to learn everything. Sometimes, the smarter approach is to build smaller skills that work well together. That’s an interesting direction Axis Robotics is exploring with its vision of a composable library of robotic capabilities. Think about a simple everyday task: organizing a kitchen table. The robot may need to recognize objects, pick them up, move them safely, and place each one in the right position. These are different capabilities, but they all need to work together to complete the task. Here’s why this approach matters: → Build skills individually. Train specialized capabilities for specific tasks instead of expecting one model to master everything at once. → Connect the skills. The output of one capability can become the starting point for another, allowing the robot to handle longer sequences of actions. → Reuse what works. A useful skill could become part of many different workflows rather than being limited to a single demonstration. → Keep improving. As more tasks are solved and tested, those capabilities can help support stronger and more general-purpose robotic models. In its September research update, Axis reported experiments where specialized “Expert” policies could be chained together to perform longer-horizon behaviors. The bigger idea is to move beyond collecting more individual tasks and toward building a reusable library of robotic skills. That’s the part I find interesting. The future of Physical AI may not depend only on how much data we collect, but also on how effectively learned capabilities can be combined and reused. Axis Robotics is exploring that direction, and I’m curious to see how far this approach can go. @axisrobotics @KaitoAI #AxisRobotics #PhysicalAI

Good morning, Web3 fam ☀️💚 Exploring the future of onchain finance with @Canborsa_DEX bringing crypto and real-world assets together in one platform. 🌐 One vision, multiple markets, and a new chapter for Web3. 🚀 #Canborsa #RWA #Web3

One thing that stands out to me about @axisrobotics is how it approaches the data problem in Physical AI. Collecting more data is important, but collecting the right data can be even more valuable. As a robot model improves, some tasks become easier while other situations still expose its weaknesses. Axis is building a system where those gaps help guide what data should be collected next. The process keeps evolving: Collect → Train → Evaluate → Improve the Data → Train Again. I like this approach because it treats robot learning as an ongoing process, not a one-time training step. And with community contributors helping generate demonstrations and corrections, the data engine can keep adapting as models become more capable. The future of Physical AI won't depend on models alone. It will also depend on how intelligently we help them learn. Curious to see how far this model-guided data approach can go. 🤖 @KaitoAI #AxisRobotics

One thing I find interesting about @axisrobotics is that its approach to Physical AI goes beyond simply collecting more data. With Axis V2, the idea is to create a continuous feedback loop. A robot policy attempts a task, its weaknesses help identify what needs improvement, and human corrections become useful training data for the next iteration. That changes the way we think about robot learning. Instead of treating data collection as a one-time process, the system is designed to keep learning from what goes wrong. And that’s where community contributions can make a real difference. Building smarter robots isn’t only about better models. It’s also about giving those models the right experience to learn from. Curious to see how far this feedback-driven approach can take Physical AI. @KaitoAI #AxisRobotics

Most people talk about how intelligent AI models are becoming. But when it comes to robotics, there’s another challenge that deserves just as much attention: where does the data come from? A robot needs more than a powerful model. It needs diverse, high-quality examples of how to interact with the physical world. That’s what makes @axisrobotics interesting to me. Its approach focuses on building a compounding data engine for Physical AI, where human contributions can help create the training data needed to improve robotic capabilities. And the idea of connecting this kind of infrastructure with a broader creator ecosystem through @KaitoAI makes the journey even more interesting. We’re still early in Physical AI, and I’m curious to see how this space evolves. The real question isn’t just how smart robots can become, but how we help them learn. #AxisRobotics