DatraAI

GEN-1.5 in-context learning is test-time training done right. Physical prompting—3–12s demos packed into a 30s context at 100Hz—gets 59% one-shot success across 10 short-horizon tasks; 10 updates on ~5 min / 50 demos reach 83%, while weights move <0.15%. No meta-objectives, no architecture tricks. Scale on interaction data made ICL appear. Conceptually, this is TTT without conventional gradient updates. Future is pretraining without action labels, but with long-horizon physical interaction data. That may be the recipe that breaks today’s action-label bottleneck. Sanja Fidler (Veeda AI) is building the world-model substrate for it. 1️⃣ Hardware isn’t the bottleneck. Data is. Stanford PR1 teleop already showed in 2008 that the hardware could do impressive things. Missing ingredient = scalable execution data. 2️⃣ “Labels can be re-done; collection cannot.” Annotation is revisable. Missed sensor signals are permanent. Episode count is rising faster than diversity—we will regret the homogeneity in five years. 3️⃣ Ego video ≠ cheap action labels. 30 Hz RGB never recovers contact forces. Its real value is learning behavioral consequences. Pseudo-labeling is looking for keys under the streetlight. 4️⃣ Not a data war → data buffet. Ego = cheap high-volume appetizer. Dexterity needs real-robot or UMI-class data. Everyone forgot simulation. Naveen’s mix: sim large, ego medium, teleop small. Simulation is the main course. 5️⃣ The real long tail is recovery. Factory ego data is the head, not the tail. True long-tail = recovering from bizarre states humans never enter. That data only appears when you actually run the robots. 6️⃣ Post-training never disappears. Motor wear + environmental drift create perpetual distribution shift. Adaptation becomes lighter and more in-context, but the residual never reaches zero. If given 100M hours: Colleen (@GeneralistAI): maximize diversity of completion methods. At that scale the model can become modality-agnostic. Ashwin (@physical_int): fill gaps—real-robot for manipulation primitives, human data for broad physical common sense, UMI for scenes robots cannot easily enter. Naveen (@FlexionAI): sim-dominant, ego medium, teleop light. Philip (@xdofai): maximize embodiment diversity—teleop 50% / human 25% / UMI 25%. Humans are another embodiment; UMI is a human with a different end-effector. Hardware was ready in 2008. Action labels may be the next bottleneck to disappear. The real scaling variable is physical experience × diversity × embodiment × horizon.

Here is why you can't sell your robotics data. Everyone admit that the momentum for robotics data is real. But every emerging market deserves a closer look beyond the hype. One structural challenge in embodied AI data is rarely discussed: The buyers and suppliers of data are often funded by the same pool of capital. Look at today’s customers of data collection companies: foundation model teams, robotics startups, traditional robot manufacturers entering AI. Most are still pre-revenue or early in commercialization. Their data budgets are often funded by recent venture financing rather than operating cash flow. This creates a fragile dynamic. A robotics company raises capital, then invests in data collection, then uses progress to raise the next round, then continues spending on more data. In this environment, data companies are not only serving market demand, they are indirectly participating in the capital cycle of the robotics industry. That means the market remains highly correlated with the success of downstream robotics companies. If robot deployments fail to reach meaningful commercial scale, data demand could slow down at the same time as funding conditions tighten. The long-term winners in embodied AI data will likely not simply be companies that collect more hours of demonstrations. They will likely be either Infrastructure platforms or Data intelligence companies that solve the long-tail challenges of real-world deployment.

Building Actuate — DatraAI's multimodal data pipeline for physical AI. QA, Synced egocentric video + IMU + hand-pose, UMI gripper data from real household and industrial environments, processed into training-ready datasets for VLA and robotics foundation models. If your team is stuck turning raw multimodal data into training-ready data for VLA, let's talk.

Most robot teleoperation copies hand motion but not "feel" so the robot crushes or drops whatever it's holding, and you throw the demo away. A new @NTUsg paper (from Research Fellow Haichao Liu, Dr. Ziwei Wang & team) addresses this: the robot reads its own fingertip tactile feedback and auto-corrects grip force in real time, while the human stays in control of the motion. Result: success on a hard assembly task went from 4% → 97%, using less than half the force. The bottleneck in dexterous robots training is also about collecting clean data where nothing got crushed along the way. This research help closing the gap. So do the #SharpaWave, that enable high quality data collection for complex manipulation tasks, like screwing. Link to the paper in comments! #DexterousManipulation #TactileSensing #AIRobotics #NTU #Dexteroushands #Robots #Teleoperation

Who & when will get to the “GPT moment”? My definition is: 40% success rate, any task, anywhere. My bet is UMI-like will get there first, given the current investment and technology. Ego will follow soon after. Teleop may never, not because of infeasibility, but because of impatience. 5/

It’s a rare privilege to be bitter-lesson-pilled in robotics. You often hear voices in robotics saying “scaling up X-type data won’t solve Y”, and without fail, the folks saying this have not actually tried scaling up the data. I don’t blame them. It takes an incredible amount of work to get high-quality data at scale. Once you have this, though, it’s amazing how it helps cut through the noise and reveal the research problems that matter. The research Sunday does would not be possible at all without the extraordinary efforts of our 1,000 Memory Developers and our elite Ops team. I’m grateful that I get to help build on their work.

Your robot can now see in depth 👀📏 LeRobot v0.6.0 adds real end-to-end depth support. Plug in an Intel RealSense, set use_depth: true, and depth maps join your RGB observations automatically: captured in millimeters, compressed into compact 12-bit depth video alongside your normal camera streams, and decoded back to physical units at training time. Depth renders live while you record and in the dataset viewer, and it works across SO-100/101, Koch, OpenArm, reBot, Unitree G1 and more. You also get full control over how everything gets encoded, RGB or depth, codec, quality, presets, all of it. More on how it works: huggingface.co/docs/lerobot/…

A Stargate for Data Labs are on a trajectory towards >$100B/year of data spend by 2030. As we begin the trillion-dollar compute project, we need to think about the equivalent civilizational-scale effort for the other core ingredient: data. At the foundation of the scaling revolution is a simple empirical law: deep neural networks improve smoothly, near magically, as you scale two things in proportion — (1) the size of the model and (2) the amount of data you train on. And despite the scaling laws being brutally diminishing, we’ve successfully bitten the bullet of logarithmic scaling with exponentially larger clusters and datasets, and received incredible new capabilities in return. But this exponential scaling is bound to hit some limits. Oddly enough, compute has compounded fairly smoothly without limit, with trillions flowing into hypercluster buildout. Instead, we’re starting to hit the limits of an exponential demand for data. Gone are the days of being purely in the compute-limited regime, where we had effectively infinite internet data but never enough GPUs, we’re now entering a data-limited regime. Luckily, this limitation is coinciding with staggering improvements in AI capabilities. Incredibly, we seem to have a real line of sight towards automating a majority of knowledge work with the methods we have today. RL + pretraining, and the data for each, will be generally sufficient to achieve most economically valuable tasks, given some minimal algorithmic progress and continued compute scaling. In a data-limited world, economic progress & scientific acceleration will be directly bottlenecked by our coverage in each domain. We need to see data collection as imperative, deserving the same civilizational ambition we’ve given compute. The internet as a one-time subsidy It’s underrated how much all progress in AI owes everything to the blessing of the internet, this one-time civilizational subsidy to deep learning, decades of unintentional accumulation of a perfect dataset: every book, blog post, image, video, paper, discussion, etc. all digitized and freely available. Without the internet, we’d likely see comparably minimal progress in AI today, and in fact, if you notice where systems currently underperform, it’s almost always a domain where web coverage is limited and data is private, expensive, non-digitized, or non-existent. But we’re running out of it. There are only about 300 trillion tokens of useful public human text, and the internet doesn’t produce nearly enough new high-quality data to match what scaling demands — we’re soon to hit the limits of public data for pretraining. And though the advent of RL bought us reprieve — chain-of-thought RL needed a new form of untapped data, gradable math & coding tasks, also available online — we’re quickly running dry of hard tasks for RL as well. Why do we need so much data anyways? Humans learn comparably in far less time, needing just one textbook where language models might need the equivalent of hundreds to learn a new topic. It’s possible we discover methods that are massively more data efficient — synthetic data, data efficient architectures, other exotic algorithms — but fundamental progress is slow and highly unpredictable, and the recipe we have just works today. And, while I’m wary of getting too deep here, even arbitrary data efficiency can’t replace data that just doesn’t exist in the first place. There’s a massive amount of missing information on the web: the dark matter of the internet — tacit knowledge, undocumented processes, etc. — most of which was never published and lives only inside organizations, the physical world, or just in people’s heads. I’ll leave it here and say, for reasons far longer than I can fit in this post [1], it’s best to operate on the assumption that our insatiable desire for data will continue as it has for the last decade. There will be >$100B/year in data spend by 2030 We’re not screwed yet, of course. Only a fraction of useful data in the world is on the public internet, the rest is stored inside private datasets, corporations, personal archives, universities, governments, and otherwise. Labs can and will continue to license these private datasets, or create them from scratch, like Anthropic’s book scanning project. And we’ll increasingly task human experts to manufacture new high-quality data, with a large fraction of hard RL training tasks already being sourced this way. But collecting this data, unlike before, will be expensive. As the free internet dries up and demand for data rises, we should see labs investing equally in data as compute, likely spending a significant fraction of their compute budgets on data. As we see trillions spent on compute, we should also expect hundreds of billions spent on data (human data & collection budgets), given their equivalent importance. And, notably, data spend is already tracking this way: total data spend across vendors, not counting internal lab efforts, is already roughly $7 billion per year. It’s quite reasonable we’ll see >10x by 2030. Data is the moat Data becoming increasingly private will also majorly shift the competitive landscape. While compute is a commodity — everyone buys the same chips and builds the same clusters — data really isn’t. The big reason why frontier models have felt eerily similar to one another, until now, is they were trained on substantially the same internet (pretraining data variability across labs seems pretty low). As labs diverge onto more exclusive, manually collected corpora, I think models will begin to increasingly diverge. OpenAI pulling ahead in mathematics and Anthropic in cybersecurity isn’t an accident. I really think laser-focused collection of high-quality midtraining tokens, custom RL tasks, environments, with dedicated research effort, has driven much of the visible progress in the last year. James Betker has an excellent blog about “the ‘it’ in a model is the dataset”: model architecture and compute buy you efficiency and order-of-magnitude performance, but ultimately, models, of any architecture, are such incredible approximators of their dataset that the core meat of a model boils down to just that, nothing else. Data is a major moat. AGI long, ASI short As I’ve tweeted before, I’m confident that, despite the narrative, the data labeling industry will continue to fuel great businesses and be an excellent AGI long, ASI short. The argument is just: By the time the AGI labs no longer need data, it’s probably over for everything else too [2]. In this frame, the last companies left should be the data companies, as the last speck of economically relevant data is sucked in. And these companies are already among some of the fastest-growing companies in history: Mercor, founded three years ago, is rumored to be doing $2 billion in revenue with something like a few million expert labelers under contract. While these businesses are very non-stationary, what type of data is needed shifts constantly, I don’t think that diminishes their value. The long-tail of the economy is long, and the value isn’t diminishing as you extend farther into more obscure information: as models get more capable, the value of the marginal dataset goes up, not down. Automating a full job means covering its full distribution of tasks, tools, edge-cases, and long-horizon loops. There’s some O-ring logic to it: a dataset that buys a 1% bump can justify a previously unjustifiable collection cost when it’s the difference between a system that does 99% of a job and one that does all of it [3]. The competitive dynamics of the data industry are still evolving but as demand for data is increasingly niche, ultra high-quality, expert-generated, I think we’ll see real consolidation. Again, contra-narrative, we’ll probably see true competitive differentiation built on brand, quality control of data (which, from personal experience, can vary massively), as well as in network effects from the talent networks themselves over time. We’ve already seen rapidly shifting data type demand work in favor of incumbents, benefiting those with early knowledge of where the market is headed. The binding constraint It’s truly remarkable that we seem to have the recipe — pretraining + RL — to absorb most economically valuable work, despite being far from a lot of what we expected from “AGI”. The same way chess engines revealed we never needed general intelligence to solve chess, as we originally thought, we’ll soon realize that software, mathematics, and the vast majority of the economy (including physical, just running ~3 years behind!) are the same. If recursive self-improvement or some other algorithmic breakthrough arrives, that’s wonderful, but we really don’t have to wait for it. The binding constraint between here and an automated economy isn’t that, it’s data coverage: every app, workflow, edge case, process, etc. sitting in private stores or someone’s head. Ultimately, while we make tremendous strides in more efficient model architectures, and clusters like Stargate equip us with zettaflop-scale compute, we really aren’t making rapid progress collecting the data we lack. We’ll soon live in a world where we have the methods & compute to accelerate scientific progress or economic growth, but not the data. And we’re already there today: frontier models would surely be as good at accounting/many medical tasks/legal advice as they are at software engineering if we only had the same pretraining & RL coverage as we did for code. I really want to drill this in: The speed at which we automate the economy is going to be directly rate-limited by our ability to collect data about it. Worth noting that under this assumption, with data as defensible and directly proportional to economic & scientific progress, data should also be considered a national strategic asset like compute. Imagine what we’d do in a world where we had a Manhattan Project-effort for AI and needed to mobilize data collection as a limiting factor. We should be concerned about China, with greater state capacity and authoritarian economic control, being capable of mobilizing data collection at national scale, potentially compounding their economy and scientific output faster than us down the line. A Stargate for data I’m leaving my complete ideas for a future post, as this one is already far too long, so I’d really like to pose the question here. Stargate exists because we organized trillions of dollars, international strategy, gigawatts around compute as a fundamental ingredient. What would equivalent ambition look like for data? Obviously, scaling data collection, a heterogeneous mass of information across the economy, isn’t going to be as clear as scaling compute, as a homogenous infrastructural effort. A core division will be first, coverage — all uncaptured knowledge sitting across the economy/science/physical world and all that simply isn’t recorded — and, secondly, sheer volume in the domains we already train on: more hard math tasks, more high-quality web text, way more coding data, more legal drafts, etc. I have a post coming soon which breaks down my proposals. There’s a lot of room for creativity. Quickly, we’ll probably want to start with a deep census of what we have and what we’re missing, predict what the 2030 model will still be bad at and work backward to what we should be collecting today. You can probably license a large amount, leveraging high lab valuations to buy datasets or companies altogether. There’s an adversarial nature to a lot of this collection with firms, so there’s lots of engineering to do this correctly. We should go convince important companies to turn off deletion policies, even if we’re not buying from them yet. Data flywheels in consumer products will be massive. Confidential training, government legislation for grant-funded research, running companies at a loss for their data, etc. We’re headed towards hundreds of billions in expenditure, national prioritization, and major data limitation on the horizon. We have a great opportunity to think creatively about what a megaproject for data would look like: How do we, deliberately this time, construct the next internet’s worth of data? Footnotes: [1]: I’ll probably soon publish my much longer post explaining my position on data efficiency and why the value of this data is still pretty high in most worlds regardless of new algorithms. [2]: The “AGI freeroll” bet: heads you win, tails ASI flips the world upside down anyways. [3]: We already see a glint of validation of this point, given the data market is strongly tilting towards ultra-high-quality agentic data, rather than unskilled labeling — niche expert workflows, live environments, and evaluations requiring increasingly obscure talent & knowledge — yet shows increasing, not decreasing, revenues.

A month ago I was talking to @xuster in IITK and asked him: "Everyone is collecting robotics data. Who's actually going to win?" His answer was simple: Either build the highest quality datasets with the best environments and hardware, or build the most diverse datasets. And this is why we got bullish on diversity at @Datra_ai Most people are playing a volume game. We're betting on diversity. Over the last few weeks, we've been collecting data across factories, hospitals, workshops, and educational labs. But one thing we're particularly excited about is partnering with Engineering anf ITI colleges. Why? Because these students spend years learning practical industrial skills before entering the workforce. Welding. Wiring. Electrical assembly. Machine operation. Workshop tasks. These are exactly the kinds of skills robotics companies ultimately want to automate. If future industrial robots are going to learn from humans, it makes sense to learn from skilled humans performing real tasks. We're now collecting data across industrial environments, hospitals, and technical training labs to capture a much broader distribution of real-world human behavior. Also looking to connect with robotics labs and teams building VLAs, world models, manipulation systems, or industrial automation. I'd love to understand what you're building and what data is actually missing today.