🤖 @saturdayrobotic Robotics & World Models Reading Club #28#: Dyna-2, @JasonMa2020 (@DynaRobotics). 👥 ~300 registrations Hosts: @junfanzhu98, @aurorafeng_01, @jerryhuang01 (@RoboticsCtr), @zhen_do_ob 🔥 human video as next scaling axis for robotics?

First of all, ARR means Annual Recurring Revenue. Piloting and R&D fees don't count. Secondly, deployment can only be done with clear infrastructure support. Thirdly, what can really be deployed is very visible if you've ever done any serious deployment. The video can easily show Just call it out, be more honest, and use less hype in every aspect, whether it's the model or the deployment.

Some recent thoughts. ICL is useful. RL is useful. VLA, sim2real, world models, memory—all of these technologies can contribute to building general-purpose robots. But we should remember that a general-purpose robot is still a robot: a physical product that ultimately needs to be deployed and deliver real value to customers. Once you look at the actual challenges of deployment, it becomes clear that no single technique is enough. It requires a full system, with careful attention to detail across models, hardware, infrastructure, reliability, and operations. Even at the model level, it’s unlikely that any single idea will be the answer. VLA, RL, sim2real, world models, ICL, memory, and many other approaches will likely all play a role. The real challenge is combining them in the right way, so they complement each other and maximize the capability and utility of the overall system. So IMO, there’s no need to overhype any single technical term. This industry would benefit from staying calm and patient, and focusing on building the holistic system that can actually work reliably in the real world and create real customer value.

This is our release that I am actually most excited about because it shows and proves what truly matters. Deployment is the ultimate prize and the only reliable eval for robotics. After having worked on so many research projects and models in my career, where I saw so many new capabilities I had never seen before, there was always a question in the back of my mind: do they matter and how? At a time in robotics where the signal-to-noise ratio is so low, where there’s so many demos, models, pilots, you see everyday, the conclusion I arrived at has always been shipping our models, robots, and entire systems to real customers and proving that they actually fulfill a need for real people. Showing a demo is easy, proving that the robots create sustainable value for customers over a long period of time is extremely hard. We know it's hard, but we also know it's necessary to get right. This mindset is also why I decided to start Dyna with @Lindon_Gao and @YorkYang5050 in the first place, because our heart has been focusing on the right problems from day 1. It’s great to see the whole team’s effort culminating in this first scaled deployment release of its kind for the industry. But we are just getting started, and I am more excited about the future of the physical economy than ever before.

“Every deployment must make the next one better. A problem solved in the field should leave something reusable behind — a better model, better tooling, better infrastructure, or a more general capability. ” Worth a read!

The heroes behind deployment are often underrated. A strong, general model is not enough to get robots into production. Anyone who has really deployed knows that most of the important lessons only show up in the field — and you need serious infrastructure to capture them. At Dyna, we’ve spent a lot of time building observability, auto-labeling, feedback, and evaluation systems that turn deployment into a continuous learning loop, with humans guiding the loop where they add the most value. That loop is critical. It tells us where models actually fail, what needs to be solved fundamentally, and where our foundational research should go next — instead of patching problems one deployment at a time. Deployment isn’t just a business use case for us. It’s what powers the data and feedback flywheel behind scalable robotics. More here: There’s probably a lot more behind it than you’d expect.

As Lindon said, we don't think we have to pick a lane. Robotics is different from other industry that the robot itself is the product. And it needs to work as a whole, not just getting a brain or something to work well.

We’ve seen plenty of impressive robotics demos. But now the bar must move higher: robots must create real value in the real world — and that value must scale. For us, the real test is not whether a robot can complete a task once. It is whether customers get enough value that they want to deploy more. Robotics only matters when it solves real problems: taking repetitive, tedious, dirty, and difficult work off people’s hands while creating meaningful economic value for the businesses using it. ROI and scalability are inseparable. One successful deployment can prove customer ROI, but not a scalable product or business. If every new customer or workflow requires rebuilding the solution, the economics will never scale. You only prove that ROI can scale when the same underlying technology keeps creating value across customers, workflows, and industries — while the effort, cost, and time for each new deployment keep coming down. We think about the path very simply: Demo: “It works.” You’ve proven technical possibility. Pilot: “It works here.” You’ve proven it can work in a real environment and start creating value. Scaled deployment: “We want more.” Customers keep expanding because the ROI works, and we can keep delivering that value without rebuilding everything from scratch. Every deployment must make the next one better. A problem solved in the field should leave something reusable behind — a better model, better tooling, better infrastructure, or a more general capability. What we learn from one customer should make the next deployment easier, faster, and more reliable. This is also why research and deployment must stay tightly connected. Research expands what robots can do. Deployment tells us what actually matters, where things break, and what must be solved fundamentally — not patched case by case. You need both to build a product that can truly scale. Today, we’re sharing more of what we’ve learned across a broad range of industry partners — the successes, failures, operational challenges, and hard-earned lessons behind getting robots to create real value in production. The goal is not just to make robots work. They must create real value. That value must repeat. And it must scale.

A few months ago, I wrote about the robotics “bubble” because I felt the industry was becoming too focused on technical breakthroughs while overlooking a basic business reality. Ultimately, the technology has to create real value for customers. Today, I’m seeing growing excitement around the idea that robotics may finally be entering its “GPT-3 moment.” I really do think there’s something very real behind this shift, and I’m genuinely happy to see the progress. At Dyna, we’ve been investing heavily in teachability because we believe it’s one of the key milestones for making robotics truly scalable. But for us, deployment is the ruler. Real-world use cases tell us whether what we’re building is actually working, and often expose assumptions in the previous solution that need to be rethought. We saw this directly with Dyna-2. Techniques that worked at 100,000 hours of data often needed to change as we pushed toward 1,000,000. New bottlenecks appeared and sometimes changed what we should optimize for even at the foundational level. That’s why research and deployment have to move together. Research pushes what’s possible. Deployment tells us what actually matters and where to push next. To me, that’s what a real “GPT-3 moment” means in robotics. It is not just about better models, but about building a loop that turns technical progress into real-world value and eventually real scale. Without stepping into real deployments, that signal can be too weak, and it becomes easy to mistake progress for readiness. We’ll be sharing more very soon about how foundational research, together with everything it takes beyond the model, is turning into real deployment and business impact. Stay tuned.

Seeing a hype wave around GEN-1.5, and rightfully so. Lots of respect to Pete & Andy for executing so well. The secret is in the naturally repetitive motions in human-collected data. There're 2 main sources for such repetitions: (1) Symmetric patterns. Sorting, tidying, and assembling almost never finish in one motion. Open any assembly manual from IKEA, and you find most objects symmetrical. You drive one bolt, then its twin, then the next pair. Every {bolt A, bolt B} pair is a natural continuation in context, and the second instance is a free training signal that imitates the first ("prompt"). (2) Recovery. Humans drop things all the time, but we pick them up so fast, we don’t even notice. That reflex to fix is half of our physical competence. The key insight is to keep the failed first half instead of trimming it away. If the model consumes the full arc, fumble, catch, continue, then recovery shows up organically at test time. It's funny that in-context improvement results from *NOT* over-sanitizing your data. The other critical ingredient is UMI. I've been saying for a while that teleop will not last, and GEN-1.5 is driving the final nail in the coffin. UMI is essentially a human wearing the robot gripper to collect data directly (human → data). Teleop inserts a layer of separation: human → VR/skeletal device → robot → data, which bleeds out all the human "physical intuition". The subtle sleight of hand we perform constantly with objects, the micro-adjustments, the feel of a part snapping into place, is nearly impossible to capture when you can't feel the environment directly. Once you have enough data, many behaviors can actually be zero-shot. For example, you don't even need finetuning to pick up a novel object. The model "just knows" what to do given a similar scene in the training distribution. Whether in-context learning truly works or not also depends on how far away the test is from training. Currently, the demos are still a bit too simple to conclude. I'm cautiously optimistic. Still, it's a great day in robotics.