India's Robotics Revolution Boosts Economy and Agriculture

🔥 China Is Not Just Building Robots. It’s Building the System Behind Them 🔥 Take a look at this video — shot in China — but don’t focus on what the robots are doing. Because what you are seeing is not just #robotics progress. It’s industrialized #training. For years, the bottleneck in robotics hasn’t been hardware alone, nor even algorithms. It’s been the ability to generate and scale real-world data. Teaching a robot is fundamentally different from training a digital model. You can’t scrape the internet for physical experience. You have to build it — interaction by interaction, movement by movement. And that’s exactly where this video signals something important. What China is demonstrating here is not just a collection of capable robots, but the emergence of a pipeline, something that looks closer to an industrial process: teleoperation at scale, structured data collection, and iterative learning loops that turn human demonstrations into machine capability. This is the missing layer that has slowed down robotics for decades. And that matters. Because in the long run, robotics won’t be won by the most impressive demo. It will be won by whoever can: - collect more data - learn from it faster - and redeploy improvements at scale This is where China has a structural advantage. The combination of manufacturing depth, operational scale, and willingness to invest in data-heavy systems creates the conditions for exactly this kind of loop to emerge. What might look like “just another robot video” is - in reality - a signal of system-level capability. Of course, it’s still early. The environments are controlled, the tasks are likely constrained, and generalization remains an open challenge. These robots are not yet ready for the full variability of the real world. But that’s almost beside the point. What matters is that the machinery and the process to get there is taking shape. And once that exist — once you can reliably turn demonstrations into data, data into models, and models into deployed behavior — the pace of progress can change dramatically. When the right training infrastructure comes together, progress stops being linear. It becomes exponential. And that’s exactly what we have seen in AI. The next step will be moving from controlled demos to real-world use—first in factories and logistics, then gradually into everyday life. The real shift will happen when these systems start powering useful consumer applications, from home assistance to personal robotics, turning what we see in labs today into products people actually live with. #ChinaTech

“Construction is the graveyard of robotics.” I heard this quote somewhere, and honestly, my own experience showed me something similar. I have been thinking about why, for almost two years. The easy answer is that construction is fragmented, conservative, low-margin, and messy. But I think there is a deeper reason. Time. The older an industry is, the more time humans have had to build practices, shortcuts, rituals, tools, tolerances, and workarounds around it. That sounds like it should make the industry easier to automate. But often it does the opposite. Because those best practices were not built for robots. They were built for humans. Construction was never designed to be understood by machines. A construction site is a half-built world. The floor is uneven. The map is outdated. Materials move. People improvise. Weather changes things. Nothing is exactly where it should be. For humans, this is normal. For robots, this is chaos. This is where Moravec’s paradox becomes interesting again. We usually explain it as: What is hard for humans is easy for machines. What is easy for humans is hard for machines. But I think the more interesting question is not what is hard. It is why. My hypothesis: The older an industry is, the more of its intelligence is hidden inside human practice. And the more an industry depends on humans absorbing physical uncertainty, the harder it is to automate by specialized robots or humanoids. This also connects to something I wrote about recently: versatility versus complexity in robotics (link in comment). In structured environments, robots can win by being narrow. They do one thing. They do it well. The world is arranged around that task. But in construction, mining, agriculture, farming, and many other old physical industries, the robot is asked to become versatile almost immediately. It needs mobility, perception, manipulation, planning, recovery behavior, safety logic, and some kind of common sense. Every extra layer of versatility adds complexity. And complexity compounds. A worker does not just perform the task. They constantly repair the task. They fix missing information. Interpret unclear situations. Work around bad tools. Adjust to other people. Sense material behavior. Make hundreds of small decisions that nobody writes down. That invisible layer is often the real job. So maybe old industries like construction, mining, and agriculture are not hard to automate because they are simple or primitive. They are hard because time has "optimized" them around human improvisation. They are pre-machine-legible. Robotics succeeds more easily when the world has already been structured for it: factories, warehouses, repeatable workflows, controlled environments. So maybe the future is not just about smarter robots. Maybe it is also about making the world slightly easier for robots to understand. (Will write about how to...) Find the link to the full article below #robotics #physical_AI #humanoids

Now in China a teenagers training Robots (Unitree Robotics H1, Fourier GR-1) ih special scools (like in video). Looks not so hard but they make data labeling layer exists so robots can eventually learn to operate in the real world ($2.3B invested in 2025) Then data engineers and ML specialists process the captured streams, build datasets, train embodied AI models (AI with a physical body) [holistic.news]. This is not reserch, serious centers run as public-private partnerships: local administrations (e.g., Beijing's Shijingshan district) + robotics startups (LEJU ROBOTICS, Unitree Robotics, AGIBOT) [China Daily]. Who's backing now: Sequoia China: ~$700M Hillhouse Investment Capital: ~$550M 许启明 Venture Partners: ~$400M Tencent, Alibaba Group, Xiaomi Technology - strategic rounds Your CFO would "have questions" on this ROI 😂 Fulll negative unit economics fo 7 yearls at list 😵. Projects like this get executives fired in a traditional framework. Robot - $40K capex. Human worker - $15K/year opex. What investors are actually buying: exclusive access to motion capture, force feedback, tactile patterns. The hardware (Unitree, Leju, AgiBot) collection instrument will generate revenue too. Unlike LLMs, robots need multimodal data (video, kinematics, haptics) that can't be scraped from the internet. It has to be physically generated [restofworld.org].  The bet isn't that robots outperform humans today - it's that the data they collect now will enable the next generation of AI capable of real-world generalization in 5-7 years. I shure #ROI exist via #RaaS in case. Robot-as-a-Service - fixed monthly fee + pay-per-task. Lowers entry barrier for customers, shifts operational risk to the provider. Investors get recurring cash flow plus exclusive access to the data the robot generates. Dual monetization: service revenue + data rights [Chinese Government Portal]. Within 36-48 months, androids (Tesla Optimus, Xiaomi CyberOne) will appear in logistics, elderly care, retail. Mass adoption will shift not just economics but everyday psychology - from trust in machines to another redefinition of "work". If it feels you unsettling. That's ok 😉 Sources: Rest of World, China Daily, Euronews, Hello China Tech, IFR, Goldman Sachs Research (2025–2026).

India’s Next Tech Leap: From Code to Robots For decades, India has been synonymous with software excellence—powering global enterprises through IT services and digital innovation. But a new shift is underway. 👉 India is now moving beyond software… into robotics and deep-tech engineering. A new generation of startups is building real-world robotic solutions—from manufacturing automation to intelligent logistics. This isn’t just coding anymore; it’s end-to-end product innovation combining AI, hardware, and systems engineering. 💡 What’s changing? - Engineers are exploring hardware + AI integration - Startups are focusing on product ownership, not just services - India is positioning itself in the global deep-tech ecosystem 🌍 Why this matters This evolution signals a transition: From IT services hub ➝ Innovation-driven economy Robotics is becoming the bridge between digital intelligence and physical execution, and India is stepping confidently into that space. 📈 If this momentum continues, robotics could become for India what IT was in the 90s—a defining global strength. 🔍 The question is no longer “Can India build?” It is now “How fast can India scale deep-tech innovation?” #IndiaTech #Robotics #Innovation #DeepTech #AI #MakeInIndia #FutureOfWork #Engineering #DigitalTransformation India moves beyond software, one robot at a time - The Times of India https://lnkd.in/dmKPHBjt

A humanoid robot costs ₹15 lakh. An Indian factory worker costs ₹22,000 a month. The robot loses. For now. Here's the Western math. A warehouse worker in the US costs $45,000 a year. A humanoid robot costs $20,000. Buy the robot, pay it back in 18 months. That's why BMW is buying Figure robots. That's why Figure is now worth $39 billion. That's why Goldman Sachs calls humanoids a $38 billion market by 2035. Now the Indian math. An Indian factory worker in 2026 costs ₹18,000 to ₹32,000 a month once you add PF, ESI, and gratuity. Call it ₹3 lakh a year. A Unitree G1 humanoid at $17,990 lands in India at around ₹15 lakh. Spread over 5 years, that's also ₹3 lakh a year. Before maintenance. Before downtime. Before training your mechanic to keep it running. The robot never beats the worker on wages. Not this year. Not next year. So is the humanoid wrong for India? No. The use-case is wrong. Indian plants don't have a labour cost problem. They have a labour showing up problem — in 3 specific places: 1) Night-shift confined spaces — chemical tanks, sugar silos, pharma reactors. Workers refuse these shifts. 2) Hot-zone work — next to furnaces and reactors. PPE compliance drops by hour 3. Injury claims are brutal. 3) Monsoon waste sorting — effluent zones, slush, fumes. Plants can't staff this from July to September. These aren't jobs where a robot replaces a worker. They're jobs where the worker stopped coming. A ₹15 lakh humanoid that does one of these for 4 hours at night is the math that works. Not 200 units. 1 robot. 1 job. 1 pilot. What changes in 2028? Unit costs drop to $10,000. The motion library grows from 5 tasks to 30. More jobs open up. For now: buy for availability. Not for savings. China shipped more humanoids than the US last year. India will do neither for a while. But sooner than most Western forecasts think, India will buy 200-unit lots — for 3 jobs Indian workers stopped accepting. That's the first real humanoid market in India. Not the Figure-for-BMW story. Plant owners, CFOs, CTOs — DM me "humanoid math" for the 1-page breakdown: the 3 jobs a ₹15 lakh humanoid beats in an Indian factory today, and the 3 it doesn't (yet). #Humanoids #IndianManufacturing #AIInfrastructure #RoboticsIndia #FutureOfWork #PharmaIndia #ChemicalIndustry #SugarIndustry #IndustrialAutomation #FactoryAutomation #PlantAutomation #FMCGIndia #AutomotiveIndia

China is shipping 10x more humanoid robots than the US. The truth behind the numbers might surprise you You’ve probably seen this story this week Chinese humanoid robot startups took the top six spots in global shipment rankings for 2025 Unitree shipped 5,500 units. AgiBot just hit its 10,000th robot. Tesla missed its 5,000-unit Optimus target News goes “China is dominating. America is falling behind” But if you spend a few hours digging into the actual data, the picture isn’t as clean Only 13,317 humanoid robots were shipped globally in 2025 According to Forbes, nobody actually knows how many of those are doing real work A lot of Unitree’s sales go to universities and research labs AgiBot’s deployments are mostly factory pilots The famous Chinese robot moments, the kung fu performances, the snow patterns, the marathon runs, are PR exercises. Not paying customers Industry experts like Professor Mohd Rizal Arshad at XJTLU put it bluntly. Most of the examples people see in the media are essentially demonstrations So what is actually happening? Two countries are funding the same uncertain bet, just in completely different ways In the US, investors are treating humanoid robot startups like AI platforms They assume these companies will own the software stack that eventually runs a global robot economy In China, investors are treating them like industrial hardware companies Capital is going into manufacturing, supply chain, and aggressive production Neither side is actually deploying at meaningful commercial scale yet I watched a Shark Tank episode recently that gave me cold sweat The founder had raised close to USD 1 million from investors. None of it went into selling Almost all of it went into patents and licenses She was protecting something that wasn’t generating a single dollar What’s happening in humanoid robotics is the same pattern, bit bigger scale American capital is funding the look of platform leadership. Big valuations, real partnerships, famous brands attached The units shipped are still tiny Chinese capital is funding the look of manufacturing dominance. Big production numbers, top shipment rankings The real customers using these robots to create value are still few If you’re a founder or investor, the same trap is happening at your scale Look honestly at where your capital is going. Is it producing real revenue, or just the appearance of progress? Patents. PR. Demos. Partnerships that look great on a deck These all feel like traction. They aren’t When the cycle turns, the companies that survive aren’t the ones with the highest valuations or production volumes They’re the ones who quietly built customer revenue while everyone else was building a story China and the US are both still in the story phase of humanoid robotics The interesting moment will come when the market realises that producing 10,000 robots and selling 10,000 robots are very different things The same is true for your business Chris

🦾 The robotics revolution is no longer a future bet — it is starting to happen now. AI is moving off the screen and into the physical world, where robots can see, move, grasp, and work alongside humans. In 2025, robotics startups have already attracted just over $6 billion in funding, putting the sector on track to exceed last year’s levels. Across the broader landscape, more than 1,500 robotics startups have reportedly raised around $90 billion since 2019. A few names stand out: ➡️ Figure AI: raised $675 million in its Series B in early 2024 at a $2.6 billion valuation, and later surpassed $1 billion in committed capital with a reported $39 billion post-money valuation. Its robots have also moved into real manufacturing work at BMW. ➡️ Agility Robotics: reportedly securing a $400 million round at a $1.75 billion valuation, with nearly 100 Digit robots already deployed across customers like Amazon, GXO, and Spanx. The industrial robotics market is already tens of billions of dollars today, but the bigger opportunity may be humanoids. As AI moves from software into physical machines, the addressable market expands from factory automation into labor replacement across manufacturing, logistics, retail, and services. Including humanoids, the robotics TAM becomes much larger than the traditional industrial robot market. 📈 Industrial robotics today: roughly $35 billion to $75 billion depending on the source and definition. 📈 Humanoid robotics / embodied AI: still early, but many public estimates put it in the tens of billions by the mid-2030s. 📈 Broader robotics and labor automation: can expand into the hundreds of billions as humanoids move from pilots into warehouses, factories, and services. 📢 The pattern is clear: the next wave is not just AI software, but AI embodied in machines. The companies that win will not only build smarter models — they will turn intelligence into action, task by task, in the real world. For investors who have spent the last years focused on AI, robotics may deserve a fresh look. Not as a replacement for AI, but as its next major frontier. ✏️ Contact DSM Capital for more info on how to participate as an investor. #AI #Robotics #HumanoidRobots #FigureAI #AgilityRobotics #Tesla #Optimus #MachineLearning #FutureOfWork #Automation #DeepTech #Startups #VentureCapital #Innovation

🧩 LeetCode 3661 : Maximum Walls Destroyed by Robots 📈 Difficulty: Hard 👀 At first glance this feels like a simple range problem… Each robot can shoot left or right, destroying all walls within its distance. So the obvious idea might be: “Just compute how many walls each robot can hit in each direction.” But there’s a catch that makes the problem much more subtle. ⚠️ Robots block bullets. If a bullet meets another robot before reaching a wall, it stops immediately. Which means a wall between two robots might be reachable by: the left robot the right robot both And if both robots can destroy it… we must make sure we don’t double count it. That’s where the real challenge begins. 🧠 Key Insight Instead of thinking robot-by-robot, think wall-by-wall while sweeping across robots. After sorting robots and walls: Every wall between two robots falls into one of three categories: 1️⃣ Destroyable only by the previous robot 2️⃣ Destroyable only by the current robot 3️⃣ Destroyable by both robots The third category is the tricky one. If both robots can destroy a wall, our choice of shooting direction determines who actually destroys it. So we must keep track of decisions carefully. ⚙️ The Smart DP Trick While scanning robots from left → right, we maintain two states: L → maximum walls destroyed if the current robot shoots LEFT R → maximum walls destroyed if the current robot shoots RIGHT Now suppose between two robots we found: next walls reachable only by the current robot common walls reachable by both robots Then the transitions become: If the current robot shoots LEFT It can destroy the next walls. But if the previous robot already destroyed the common walls, we must avoid counting them twice. So we adjust the previous result before adding the new walls. If the current robot shoots RIGHT It doesn't immediately destroy new walls (those will appear later in the sweep). So it simply inherits the best previous result. 💡 Why This Works This transforms the problem into a linear sweep with dynamic decisions. Instead of trying every direction combination (which would be exponential), we only keep two states per robot. And we update them as we encounter walls. So we solve a seemingly complex geometric problem using: ✔ Sorting ✔ Sweep line ✔ Dynamic programming ⏱ Complexity Sorting: Robots → O(n log n) Walls → O(m log m) Sweep processing: O(n + m) Final complexity: O(n log n + m log m) Which works efficiently even when both arrays are up to 100k elements. 🧩 Takeaway The trick to solving hard problems is often changing the perspective. Instead of asking: “Which robot destroys which walls?” Ask: “While sweeping the line, how do wall ownership decisions evolve?” Once you see that, the problem turns into a beautiful DP sweep problem instead of a messy simulation. And those are always the most satisfying ones to solve. 🔥 #LeetCode #Algorithms #CompetitiveProgramming #DataStructures #DynamicProgramming #LearningInPublic #ProblemSolving

The most important robotics story of 2026 isn't happening in a German auto plant or a Silicon Valley warehouse. It's happening in Southeast Asia, where a company called Dwbrobot has quietly detonated the assumption that industrial automation requires capital. The zero-investment robot model — deploy now, pay through operations — sounds like a leasing arrangement. It isn't. It's a structural attack on the single biggest barrier keeping small and mid-sized manufacturers in the Philippines, Vietnam, Indonesia, and across the developing world locked out of the automation economy. Capital cost has always been the moat that separated automated factories from manual ones. Dwbrobot just filled that moat. Here's what conventional analysis misses: this isn't primarily a story about robots. It's a story about who gets to participate in the next wave of industrial productivity. For the past decade, automation has compounded advantage for those who already had it — large manufacturers in wealthy nations absorbed productivity gains, while smaller competitors in cost-sensitive markets doubled down on cheap human labor as their only competitive strategy. That asymmetry was accelerating. A garment factory in Cebu or a furniture maker in Pampanga couldn't access the same productivity multipliers as a factory in Stuttgart or Shenzhen. The gap wasn't closing. It was widening at the speed of capital. The zero-investment model breaks that loop. When the cost of entry drops to near zero, the question shifts from 'can we afford to automate?' to 'what do we automate first?' That is a civilizationally different question. It moves automation from a balance-sheet decision to an operational one — and operational decisions happen faster, get iterated faster, and scale faster than capital allocation decisions ever do. Policy makers in Manila, Jakarta, and Kuala Lumpur should be watching this closely — not to regulate it, but to accelerate it. The ASEAN Secretariat and national industrial development agencies need to ask a pointed question: if zero-investment robotics is now technically possible, why is it not yet policy-supported? Procurement frameworks, SME development funds, and industrial park incentives are all still calibrated for a world where robots cost millions upfront. That world ended. The frameworks haven't caught up. The deeper challenge is workforce transition. Zero-investment deployment means the speed of adoption will surprise everyone — including the workers on those factory floors. The Philippines' Technical Education and Skills Development Authority needs a robotics co-existence curriculum deployed in 2026, not 2028.

Exciting results published by Physical Intelligence. The robots themselves still look a bit clunky, and this is obviously not “general household labor” yet. The important point is that training may be getting more efficient. If robots can learn new tasks by remixing existing knowledge, more like people do, that has the potential to accelerate a wide range of downstream use cases. Relatedly, it may also be a warning sign for startups whose whole pitch is selling robot datasets. We wrote a bit about that broader setup in our latest robotics piece. Link to Physical Research's actual paper and our recent robotics post in comments. https://lnkd.in/ggH98umN