Rapid Prototyping Techniques

Exciting updates on Project GR00T! We discover a systematic way to scale up robot data, tackling the most painful pain point in robotics. The idea is simple: human collects demonstration on a real robot, and we multiply that data 1000x or more in simulation. Let’s break it down: 1. We use Apple Vision Pro (yes!!) to give the human operator first person control of the humanoid. Vision Pro parses human hand pose and retargets the motion to the robot hand, all in real time. From the human’s point of view, they are immersed in another body like the Avatar. Teleoperation is slow and time-consuming, but we can afford to collect a small amount of data.  2. We use RoboCasa, a generative simulation framework, to multiply the demonstration data by varying the visual appearance and layout of the environment. In Jensen’s keynote video below, the humanoid is now placing the cup in hundreds of kitchens with a huge diversity of textures, furniture, and object placement. We only have 1 physical kitchen at the GEAR Lab in NVIDIA HQ, but we can conjure up infinite ones in simulation. 3. Finally, we apply MimicGen, a technique to multiply the above data even more by varying the *motion* of the robot. MimicGen generates vast number of new action trajectories based on the original human data, and filters out failed ones (e.g. those that drop the cup) to form a much larger dataset. To sum up, given 1 human trajectory with Vision Pro  -> RoboCasa produces N (varying visuals)  -> MimicGen further augments to NxM (varying motions). This is the way to trade compute for expensive human data by GPU-accelerated simulation. A while ago, I mentioned that teleoperation is fundamentally not scalable, because we are always limited by 24 hrs/robot/day in the world of atoms. Our new GR00T synthetic data pipeline breaks this barrier in the world of bits. Scaling has been so much fun for LLMs, and it's finally our turn to have fun in robotics! We are creating tools to enable everyone in the ecosystem to scale up with us: - RoboCasa: our generative simulation framework (Yuke Zhu). It's fully open-source! Here you go: http://robocasa.ai - MimicGen: our generative action framework (Ajay Mandlekar). The code is open-source for robot arms, but we will have another version for humanoid and 5-finger hands: https://lnkd.in/gsRArQXy - We are building a state-of-the-art Apple Vision Pro -> humanoid robot "Avatar" stack. Xiaolong Wang group’s open-source libraries laid the foundation: https://lnkd.in/gUYye7yt - Watch Jensen's keynote yesterday. He cannot hide his excitement about Project GR00T and robot foundation models! https://lnkd.in/g3hZteCG Finally, GEAR lab is hiring! We want the best roboticists in the world to join us on this moon-landing mission to solve physical AGI: https://lnkd.in/gTancpNK

What if the best solutions for your process started with cardboard? When testing new ideas or improvements, jumping straight to high-cost, permanent solutions can be risky—and expensive. That’s where cardboard engineering comes in. Cardboard is one of the simplest, most cost-effective tools for rapid prototyping and testing ideas. It’s lightweight, easy to shape, and lets you visualize, test, and refine your concepts before committing to more expensive materials. Why Cardboard Is Perfect for Prototyping: 1️⃣ Low-Cost Experimentation Testing with cardboard lets you try multiple iterations of a design without worrying about material costs. 2️⃣ Fast Feedback Loops You can build and modify a prototype in minutes, gathering instant feedback from your team or operators. 3️⃣ Hands-On Collaboration Cardboard prototypes allow teams to actively engage with ideas, making it easier to identify issues or opportunities for improvement. 4️⃣ Visual Validation Sometimes, seeing a physical model highlights challenges that wouldn’t be obvious in a drawing or plan. How to Use Cardboard for Lean Improvements: 🔍 Test Workstation Layouts Use cardboard cutouts to mock up layouts and placement of tools, parts, and equipment. Adjust until everything flows smoothly. 📦 Simulate Material Flow Prototype racks, bins, or carts to ensure materials are stored and moved efficiently before building them with more durable materials. 🛠️ Design Fixtures or Jigs Create cardboard versions of fixtures or jigs to test their functionality in the process. Refine the design before investing in the final version. 📐 Test Ergonomics Mock up equipment or workstation designs with cardboard to test ease of use, reach, and operator comfort. Example of Cardboard in Action: A manufacturing team wanted to redesign a workstation to reduce operator motion. Instead of committing to expensive reconfigurations, they used cardboard to prototype the layout. After several iterations, they found the optimal setup, reducing motion by 25% and saving hours of work. Cardboard isn’t just for packaging—it’s a powerful tool for testing and refining your ideas. By prototyping with low-cost materials, you can experiment, learn, and improve quickly without breaking the bank.

Aussie experimentation programs should 10x test volume this year! Here’s why and how 👇 I’ve spoken to multiple teams running experimentation/CRO/AB Testing programs in Australia. Most run 1 test per week to 1 per month (or 12 to 52 tests per year). For comparison, world-class companies where I built experiment platforms run hundreds to thousands of tests concurrently, every single day. Now, imagine running thousands of tests per year. Even if 90% were inconclusive (which is normal), you’d still learn far more than a company running only a handful. Most companies don’t need THAT kind of volume. Going from 12 to 120 or 52 to 520, however, is absolutely doable and would be a game-changer. So, what’s stopping teams from scaling experimentation? Here are 4 common blockers and how to tackle them: 1️⃣ 𝗙𝗲𝗮𝗿 𝗼𝗳 𝘁𝗲𝘀𝘁 𝗶𝗻𝘁𝗲𝗿𝗮𝗰𝘁𝗶𝗼𝗻𝘀 • Teams avoid overlapping experiments, fearing interference. • But (a) research shows interactions are rare, (b) they’re easy to detect and manage, (c) you want to find them if they exist! • Avoiding overlaps means leaving hundreds of tests and massive insights on the table. 2️⃣ 𝗟𝗶𝗺𝗶𝘁𝗲𝗱 𝗶𝗱𝗲𝗮𝘁𝗶𝗼𝗻 • Teams struggle to generate enough test ideas. • This is often due to lack of exposure to large experimentation programs and creative testing methods. • My advice: ✅ Expand your test scope and go beyond UI to UX, pricing, marketing, and customer segments. ✅ Set up a new stream of strategic experiments, not just tactical ones. 3️⃣ 𝗘𝘅𝗲𝗰𝘂𝘁𝗶𝗼𝗻 𝗰𝗼𝗻𝘀𝘁𝗿𝗮𝗶𝗻𝘁𝘀 • Limited engineering time or bandwidth slows teams down. • Three ways to fix this: ✅ Prioritise high-impact experiments: bigger swings, more strategic focus. ✅ Reduce test effort and simplify experiments to 2-5x volume without extra strain. ✅ Secure more resources and prove value, influence leadership, and push for investment. 4️⃣ 𝗩𝗲𝗻𝗱𝗼𝗿 𝗰𝗼𝘀𝘁𝘀 • Some experimentation platforms charge hefty fees as test volume scales. • If this is blocking you, explore new vendors (happy to recommend alternatives). Well, there’s my brain dump for the week. What’s your biggest challenge in scaling experimentation?

How to get your entire team to prototype with AI? This is one of the most actionable AI articles I’ve read this year—especially if you’re a PM, designer, or eng lead trying to scale team adoption. This guide by Colin Matthews changed how I think about bringing AI into the product lifecycle—not just as a tool for ideation, but as a shared team ritual. According to Colin, the biggest barrier isn't learning v0 or Bolt—it's the gap between individual experiments and collective capability. Most PMs can now spin up a decent prototype in 20 minutes. Yet teams still struggle with two critical challenges: creating prototypes that look credible enough for stakeholders, and moving beyond isolated exploration to shared workflows. The solution lies in component libraries as shared infrastructure. Think of them as your team's design DNA—reusable building blocks that maintain brand consistency while accelerating individual creativity. Three approaches emerge with different effort-to-output ratios: - Screenshots (lowest barrier): Upload UI samples, generate matching components. Works with any tool, requires zero technical expertise. - Chrome extensions (middle ground): Extract components directly from live websites. Magic Patterns leads here, offering real-time component capture. - Code integration (highest fidelity): Connect to your actual codebase for pixel-perfect prototypes. Requires engineering support but delivers production-quality results. The breakthrough happens when you combine component libraries with baseline-and-fork workflows. Build one high-quality reproduction of your current product, then fork it for each new experiment. No more rebuilding from scratch. This transforms AI prototyping from individual exploration to team capability. Everyone starts with the same quality foundation, iterations happen faster, and conversations focus on features rather than visual distractions. What's the biggest gap between your team's AI experiments and actual workflow integration? Read the original article: https://lnkd.in/efNY9HK3 — 👋 This is Shyvee Shi — former LinkedIn product leader, now building the AI Community Learning Program, powered by Microsoft Teams. If you're curious about building and upskilling with AI, you can join our AI Community and get access to curated resources, tools, programs and events via aka.ms/AICommunityProgram. ♻️ Repost to help someone learn, build, and grow in the AI era. #AI #ProductManagement #Prototyping #AIAdoption

How do you plan when uncertainty only seems to grow? Through embracing disciplined experimentation. Here’s new writing from our Partner Charlotte Desprat on the five-step process we use to make a company great at it: 1. First, establish what you know as fact and what you don’t know – including the X-factors that could upend your plans. 2. From there, tease out the key hypotheses that you want to test. Keep in mind that some hypotheses might be more fundamental than others, and therefore might need to be tested earlier. 3. Then, consider how you might investigate each of these hypotheses using the scientific method. How can you break hypotheses into small, easily-testable components? Depending on the degree of unknowns, a rapid-fire approach might be enough to determine the key components of change. 4. Once you’ve designed your experiments, consider the time, cost, and risk associated with each. Together with the importance of each hypothesis, decide which experiments must come first vs. later. This will give you a priority list to adjust along the way. 5. Finally, set up a system by which you can quickly capture learnings and adjust. Obtain tangible measurements from these experiments. Your system should include a way to decide which experiments to follow up with, know if more are needed, and determine when you’ve learned enough from a given test. Critically, it should include a mechanism to end new ideas. Remember that about 80% of venture capital investments fail, and yet venture capitalists earn higher return on capital employed than public companies; their secret is that most of their failures come early, quickly, and cheaply. By treating experimentation as a discipline, not a one-off, you can capture the upside of uncertainty. That will be one of the most important capabilities to win in a turbulent future. Interested in our book chapter on experimentation? Click here for a direct download: https://lnkd.in/eAnUrC2t

The usual thinking often goes, "We're changing the website/platform, so there's no point optimizing what we already have." This perspective, while common, can inadvertently equate experimentation solely with optimisation, potentially overlooking the enormous benefits of integrating a truly experimental approach into development and innovation. A replatforming or redesign project typically involves a complex decision-making and MoSCoW-style exercise centered around a set of features. It's often impossible to exactly replicate old features on a new platform, meaning crucial decisions must be made about what's essential and what might be dropped. Likewise, new platforms can introduce various potential new features, but are they truly worth the investment? These decisions can become complex, political, and increasingly stressful as deadlines loom. The risk is that choices are made based on internal influence rather than what will genuinely serve the customer, which is inherently difficult to guess. How can you better manage this process? How can you genuinely know what will deliver the best customer experience and commercial outcomes? EXPERIMENTATION! When done properly, experimentation (including but not limited to A/B testing) can fast-track this entire process and help you deliver a project that actually works. Consider starting by creating a comprehensive list of all feature disparities that need to be addressed. Then, establish an initial prioritization. Next, plan and run experiments for each consideration. Finally, assess the likely benefit. Some experiments are remarkably straightforward. If a new platform won't include a particular feature "out of the box," you could A/B test removing it from your existing site to understand its true importance. Others might be more challenging. If a new platform offers recommendations but at additional cost, you could conduct more rudimentary experiments on your existing site to test the core concept. Moreover, these features don't have to be front-end; the same process can be applied to backend operational features if you have the right expertise. Experimentation isn't just optimisation; it's a critical tool for informed innovation. #experimentation #cro #productmanagement #growth #digitalexperience #experimentationledgrowth #elg #growthexperimentation

The defense sector is at a pivotal moment. JCIDS has become an unwieldy beast, stifling innovation and operational agility. It's time for a bold move - as Dan Patt and Bill Greenwalt argue in Required to Fail - JCIDS is beyond redemption. They propose the Joint Operational Acceleration Pathway (JOAP), a budget-driven model focused on real-world experimentation rather than endless paperwork. Imagine: • JOAP: Instead of thick requirements docs, prioritize prototyping and user feedback. • Joint Acceleration Reserve (JAR): A strategic holdback in service budgets to fund truly joint capabilities. • Mission Engineering and Integration Activity (MEIA): A lean organization for swift project execution. The defense tech landscape is shifting with companies like Palantir Technologies and Anduril Industries challenging traditional giants. Elon's influence at the Pentagon could be the catalyst for change, emphasizing efficiency over bureaucracy. • End JCIDS: It's time to break free from bureaucratic quagmires. • Embrace Speed: Move from planning to prototyping to production at pace. • Foster Innovation: Encourage a culture where new players can thrive alongside established firms. The future of defense isn't in more docs but in arming warfighters with cutting-edge, rapidly deployable tech. Let's make this the moment we redefine how the DoD innovates. Read the latest from Matt and I on Defense Tech and Acquisition https://lnkd.in/eHqVZExg

“OTA — OTHER TRANSACTION AUTHORITY” … PART 4 OF 6, DOD TOOLS FOR RAPID ACQUISITION AND PRODUCTION: Other Transaction Authority (OTA) is a vital contracting tool that allows the Department of Defense (DoD) to rapidly acquire and develop cutting-edge technologies by bypassing some of the regulatory burdens of traditional acquisition methods. OTA fosters innovation by enabling partnerships with non-traditional contractors, such as small businesses, startups, and academia, to address critical Warfighter needs. Successful Examples of OTA in Action 1. Hypersonic Weapons Development: The DoD used OTA to streamline the prototyping and production of hypersonic glide vehicles, significantly accelerating timelines to counter peer threats. 2. Counter-Drone Systems: Rising UAS threats led the DoD to use OTA to field counter-drone technologies quickly, ensuring operational readiness in contested environments. 3. AI for Battlefield Intelligence: OTA agreements helped integrate AI and machine learning into programs like Project Maven, improving real-time decision-making on the battlefield. Future Opportunities for OTA 1. Quantum Defense Technologies: Partnering with private-sector quantum computing leaders, OTA could advance secure communications and encryption against sophisticated cyber threats. 2. Space-Based Defense Systems: OTA can drive the development of missile defense satellites and on-orbit repair technologies to sustain U.S. superiority in the space domain. 3. Soldier Enhancements: Wearable technologies like augmented reality (AR) systems and exoskeletons could be developed faster through OTA, improving Warfighter performance and survivability. 4. Bioengineering for Combat Resilience: OTA could accelerate adaptive biotech solutions to enhance soldier endurance in extreme environments, such as desert or arctic operations. 5. Climate-Adaptive Infrastructure: To ensure readiness amid climate challenges, OTA could enable rapid deployment of renewable energy solutions and resilient base infrastructure. OTAs are used for three main purposes: • Prototyping: Testing and refining new ideas or technologies. If a prototype works, it can quickly move to production. • Research and Development: Partnering with experts to explore emerging technologies like artificial intelligence and hypersonics. • Production: After successful testing, technologies can be fast-tracked for deployment. Legislative Authority: OTA’s legal foundation comes from U.S. laws, including updates from the National Defense Authorization Act (NDAA), specifically 2016 and 2018. More recent NDAAs enhanced oversight and transparency while preserving OTA’s flexibility. OTA’s flexibility allows the DoD to move at the speed of innovation, addressing urgent Warfighter needs while fostering collaboration with the private sector and academia.

Hello 👋 from the Automate Show in downtown Detroit. I’m excited to share with you what I’m learning. Robotics is undergoing a fundamental transformation, and NVIDIA is at the center of it all. I've been watching how leading manufacturers are deploying NVIDIA's Isaac platform, and the results are staggering: Universal Robotics & Machines UR15 Cobot now generates motion faster with AI. Vention is democratizing machine motion for businesses. KUKA has integrated AI directly into their controllers. But what's truly revolutionary is the approach: 1. Start with a digital twin In simulation, companies can deploy thousands of virtual robots to run experiments safely and efficiently. The majority of robotics innovation is happening in simulation right now, allowing for both single and multi-robot training before real-world deployment. 2. Implement "outside-in" perception Just as humans perceive the world from the inside out, robots need their own sensors. But the game-changer is adding "outside-in" perception - like an air traffic control system for robots. This dual approach is solving industrial automation's biggest challenges. 3. Leverage generative AI Factory operators can now use LLMs to manage operations with simple prompts: "Show me if there was a spill" or "Is the operator following the correct assembly steps?" Pegatron is already implementing this with just a single camera. They're creating an ecosystem where partners can integrate cutting-edge AI into existing systems, helping traditional manufacturers scale up through unprecedented ease of use. The most powerful insight? Just as ChatGPT reached 100 million users in 9 days, robotics adoption is about to experience its own inflection point. The barriers to entry are falling. The technology is becoming accessible even for mid-sized and smaller companies. And the future is being built in simulation before transforming our physical world. Michigan Software Labs Forbes Technology Council Fast Company Executive Board

The old adage of “measure twice, cut once” is great when you are buying expensive fabric. But it puts the incentive on getting it right the first time, which goes against best practices for innovation. The solution is get cheaper fabric – and here’s what that looks like: 🚧 Put guardrails around your R&D work to help teams feel more comfortable failing. Innovation should not look like expensive, long-drawn out experiments with no results. Expect some sort of learnings (results) in a shorter amount of time, with a tighter budget. 🔍 Expect your teams to prototype and be smart about testing assumptions. Think like a scientist and design a smart experiment that can get you answers quicker, faster, cheaper. If you don’t need silk to prove that your shirt design fits the model, then use the cheaper fabric before sinking cost. This is hard to learn as an organization, and hard to do generally because we are wired to add on more and more, “yes, and”ing into infinity until we’ve added so much bloat and scope that the costs rise high. We need to learn a lean and subtractive mindset that narrows in on the factors that indicate success, and develop smart tests that help us understand if our core assumptions make sense. Leaders can get wary of innovation work because typically we assume it’s a big money and time sink without much ROI, but we’re doing it wrong. We have to create safe spaces for teams to experiment and fail (and learn), that the organization is comfortable with the investment, and test more ideas through cheaper, faster experiments. These are smart guardrails that we can establish to cultivate a more innovation-friendly environment. #innovation #design #leadership #designthinking This is the 2nd of 3 posts this week on innovation. Connect with me if this resonates ~