The robotics boom: Why morphology-agnostic intelligence is the key

Om Malik's thoughts on AI from a conversation he had with Rodney Brooks, some MIT nerd who founded a few robotics companies. "We recently connected for a conversation about robotics, artificial intelligence, and the future. Contrary to many, he believes humans will do just fine in a world filled with robots and AI. He poured cold water on the humanoid robot hype. He also said that if you look at the computer and internet revolutions, the AI revolution is going to take a lot longer than most think. “There’s a tendency to go for the flashy demo. But the flashy demo doesn’t deal with the real environment. It’s going to have to operate in—the messy reality. That’s why it takes so long for these technologies,” he said. He painted a more pragmatic yet optimistic vision ahead. He warned that humanoid hype is creating a lot of false expectations. Excerpts from our conversation." https://lnkd.in/d2uUGrTb

Robots receive major intelligence boost thanks to Google DeepMind's 'thinking AI' - Live Science Google DeepMind has unveiled groundbreaking advancements in AI robotics with the release of Gemini Robotics models 1.5 and ER 1.5. These models enable robots to execute complex, multi-step tasks with reasoning and adaptability that were previously unattainable. From sorting fruit by color to following real-world recycling rules, these robots perform tasks while explaining their decision-making in natural language. This innovation marks a significant leap forward in AI-powered robotics, making robots capable of generalized reasoning and effective learning across multiple systems. As robots become more intuitive, their potential to impact industries and daily life grows exponentially. Learn more at: https://lnkd.in/db4tYUi6 What are your thoughts on this? Don't hesitate to share your thoughts and ideas in the comments below. DevTech is always eager to hear from our community and learn about your experiences and perspectives. Looking forward to connecting with you! #devtech.pro #AI #technology #trending #news #innovation #technology This article is written and published by Doki. Doki is our documentation's and social media's AI Agent.

🦾💸 The Humanoid Hype: Are We Funding Fantasy? Rodney Brooks, co-founder of iRobot and MIT veteran, just lobbed a truth grenade into the humanoid robotics gold rush. In his latest critique, he calls out the billions pouring into startups like Figure and Tesla’s robotics arm as “pure fantasy thinking”. Why? Because teaching robots dexterity via human video demonstrations ignores the brutal reality: human hands have ~17,000 touch receptors. Robots? Not even close. And when these bipedal bots fall, they don’t just stumble—they become safety hazards, packing exponentially more kinetic energy as they scale. Brooks predicts that the real winners won’t look human at all. Think wheels, arms, sensors—not sci-fi silhouettes. Meanwhile, today’s humanoid prototypes are expensive experiments unlikely to reach mass production. 🔍 What’s striking isn’t just the technical critique—it’s the cognitive dissonance. Investors perceive speed and progress, yet studies show AI tools can slow developers down by 19%, even as they feel 20% faster. Perception ≠ performance. 💡 For those of us navigating AI ethics, urban innovation, or digital hygiene, Brooks offers a sobering reminder: not all moonshots are worth chasing. Sometimes, the most human thing we can do is rethink what “human-like” innovation should actually look like. #AIethics #Robotics #InnovationStrategy #TechBubble #DigitalHygiene #UrbanFutures #LinkedInThinkers

Embodied AI and Robotics, especially humanoid robots, are gaining a lot of momentum this year. If you’re curious about how robot learning has evolved from traditional modular systems (perception → planning → control) to modern end-to-end models like vision-language-action (VLA), join our online session this Tuesday, November 4. I’ll be hosting Dr. Benjamin Eisner, Ph.D. in Robotics from Carnegie Mellon University, for an engaging discussion on how robots learn. We’ll cover: • The shift from modular design to end-to-end learning • The role of Model Predictive Control (MPC) — used by SpaceX and Boston Dynamics’ Atlas, and its place in the VLA era • How Imitation Learning and Reinforcement Learning can be combined for better performance • The challenge of limited training data and the idea of the data pyramid: internet, simulation, real-world data, and sim-to-real gap. We’ll also explore Dr. Ben’s perspective on where robotics is heading — and how close we are to the “household robots” we’ve long imagined. The session will be high-level and easy to follow, designed for anyone interested in how robots are learning to see, move, and think. There’s more great content lined up — check the full agenda on the event site. https://lnkd.in/gwbdG_am

In case you missed it, TechCrunch featured our Co-Founder and CTO, Rodney Brooks - continuing an AI conversation that started in 2017 when he was at MIT. While AI conversations centered around different topics, the cyclical hype trends are reflective in what we're seeing today. "Human hands are packed with about 17,000 specialized touch receptors that no robot comes close to matching. While machine learning transformed speech recognition and image processing, those breakthroughs were built on decades of existing technology for capturing the right data. 'We don’t have such a tradition for touch data,' Brooks points out. Then there’s safety. Full-sized walking humanoid robots pump massive amounts of energy into staying upright. When they fall, they’re dangerous. Physics means a robot twice the size of today’s models would pack eight times the harmful energy. Brooks predicts that in 15 years, successful “humanoid” robots will actually have wheels, multiple arms, and specialized sensors and abandon the human form." Check out the full article: https://lnkd.in/gMm8fzN8

Industrial robots used to need complete retraining for new tasks. This open-source AI model learns once, works everywhere. Physical Intelligence spent billions. INSAIT made it free. SPEAR-1 changes everything. This Bulgarian breakthrough gives robots 3D vision. They understand space like never before. The model matches billion-dollar commercial systems. But it's completely open source. Why this matters: • Startups can experiment without massive budgets • Researchers accelerate progress together • Small manufacturers access advanced robotics • Innovation happens faster across the board Robot intelligence was stuck. Each new arm meant starting over. Each environment required retraining. Now we're applying the language model approach. Massive data plus computing power equals general capabilities. What seemed impossible a year ago is happening today. The democratization of robotics has begun. Open source is leveling the playing field. Just like open language models transformed AI. What industry will benefit most from accessible robot intelligence? #OpenSource #Robotics #IndustrialAutomation 𝗦𝗼𝘂𝗿𝗰𝗲꞉ https://lnkd.in/gk7w2G98

🔥 Gemini Robotics 1.5: When AI Steps Into Our Physical World 🔥 DeepMind just pushed a breakthrough: Gemini Robotics 1.5 (and its partner model Gemini Robotics-ER 1.5) brings AI agents from purely digital domains into physical space—perceiving, planning, reasoning, and acting in real environments. 🔍 What’s New & Unprecedented Embodied Reasoning + Tool Use Gemini Robotics-ER 1.5 reasons about the physical environment, calls tools like Google Search when needed, and sketches multi-step plans. Gemini Robotics 1.5 then turns that plan into action—vision → language → motion. Multi-step Task Execution Robots can now handle more than one-off commands. Think: sorting laundry by color, packing based on weather, or classifying trash according to local rules. Cross-Robot Skill Transfer One model, multiple embodiments. Skills learned by one robot (say, a dual-arm manipulator) can be applied to others (humanoid or different form factors). Developer Access Begins Gemini Robotics-ER 1.5 is now available via the Gemini API in Google AI Studio. Gemini Robotics 1.5 is rolling out to select partners. 💡 Why It Matters for Tech & Industry From “smart assistants” to “thinking robots” The shift is real. Robots will no longer just execute—they’ll reason, adapt, and strategize. Bridging AI & Physical Reality For decades, so much AI progress stayed trapped in screens and code. Now it’s meeting the messy, unpredictable real world. Lowering barriers to physical AI development By exposing embodied reasoning and tool use via APIs, DeepMind empowers robotics builders to build faster, safer, and smarter agents. The next frontier is normalization As robots gain general-purpose capabilities, the question shifts from “Can we do it?” to “How safe, reliable, and ethical can we make it?” ⚠️ A Note on Safety & Responsibility DeepMind already emphasizes safety—embedding semantic safety checks, low-level collision avoidance, and alignment with broader AI safety principles. But as we hand more autonomy to physical agents, the stakes rise. Noise, edge cases, adversarial settings—all must be accounted for. 🧠 My Take We’re in the early days of “Thinking Robots.” Gemini Robotics 1.5 feels like the moment when AI gets legs—literally. For designers, engineers, and strategists, this means imagining systems that aren’t just smart—they move, adapt, and learn in the real world. Let’s use this moment to rethink: How do you design for embodied intelligence? Where should human oversight remain non-negotiable? What domains are ripe for safe, autonomous robotic integration (healthcare, logistics, manufacturing, homes)? What do you see as the first real-world use case (beyond labs) for these thinking robots? Drop your bet below. 👇 #AI #Robotics #EmbodiedIntelligence #GeminiRobotics #DeepMind #AgenticAI #Innovation #FutureOfWork #TechTrends

Robot fun fact: The smartest robots don’t follow orders. They learn to cooperate. Here’s why: 1. No central commander. Each robot in a swarm follows simple, local rules — like ants or bees. No one robot knows the full plan, yet the group completes complex tasks together. 2) Emergent intelligence. When these independent agents interact, patterns form. The group begins to learn — adapting, problem-solving, and even recovering from failures — without being told how. 3) Resilience through cooperation. Systems built on control often break when conditions change. But systems built on cooperation self-correct. They evolve. The inverse is true, too. The more we try to command intelligence, the less adaptable it becomes. Maybe the next breakthrough in AI won’t come from building smarter machines — but from teaching them to work together. Citation: Zheng, Z., Wang, T., Xiang, Y., Lei, X., & Peng, X. (2024). Emergence of Collective Intelligence in Swarm Robotics. https://lnkd.in/ghcxqvwn

🚀 Is this how we finally get robots to master the real world? One of the biggest hurdles in robotics has always been the "sim-to-real" gap. Training a robot in the real world is incredibly slow, expensive, and risky (imagine a robot learning to use your fine china!). Virtual simulations are safer, but often too simple or unrealistic for a robot to transfer its learned skills to our messy, unpredictable homes. That's why I was so fascinated by this new tool from MIT CSAIL, called Steerable Scene Generation. It's a genuine game-changer. In a nutshell, it uses a powerful combination of generative AI (a steered diffusion model) and the same strategic search technique that powered AlphaGo (Monte Carlo Tree Search) to build incredibly realistic and physically accurate 3D worlds. But here's the magic ✨: The AI doesn't just create random scenes. It "steers" the creation process to build logical, everyday environments. It knows that forks and plates go on a table, not through it. This allows a robot to practice complex tasks, like setting a table or stacking dishes, endlessly in a safe, scalable virtual world. This is more than just a cool simulation. It's a potential solution to the data bottleneck that has held back the development of general-purpose robots. By making training faster, cheaper, and far more realistic, we're taking a massive leap toward robots that can finally navigate and interact with our world safely and effectively. To me, this feels like a foundational piece of the puzzle falling into place. We're not just building robots; we're building intelligent, dynamic worlds to teach them. What's the first task you'd want a household robot to learn? I'd love to hear your thoughts! Credits: MIT CSAIL Nicholas Ezra Pfaff #AI #Robotics #Innovation #MIT #CSAIL #MachineLearning #GenerativeAI #FutureOfTech #ArtificialIntelligence

This thread on X by Jascha Sohl-Dickstein (Member of Technical Staff at Anthropic) is a must-read for any robotics PhD student. It's a sobering and incredibly motivating take on planning a research career on the cusp of AGI. As someone deep in the world of robot perception, his arguments hit particularly close to home. The core idea—that AI will soon outperform humans on all intellectual tasks—forces a rethink of what we do in the lab every day. A few points reframed for our field: 🤖 The Perception Pipeline Problem: Imagine spending four years perfecting a multi-modal sensor fusion algorithm for autonomous navigation, only for a future Vision-Language-Model (VLM) to achieve superior zero-shot performance with a simple prompt. Sohl-Dickstein's "bitter lesson" suggests that hand-crafting complex pipelines is becoming an increasingly fragile research strategy. The real value is in work that scales with intelligence, not work that will be replaced by it. 🧭 Beyond Benchmarks: If the "safe" path of chasing incremental gains on established datasets like KITTI or nuScenes is no longer truly safe, it liberates us to tackle weirder, more fundamental problems. What about perception systems that learn from true physical interaction and embodiment, not just static datasets? Or developing the principles for verifiable, robust systems that we can actually trust in safety-critical applications? Perhaps our unique advantage lies in creating unique, embodied data that simply cannot be scraped from the internet. 🔧 Leverage Through Embodiment: It's easy to feel like a small cog, but the talk highlights our immense leverage at this early stage of the exponential. For us in robotics, that leverage isn't just in algorithms, but in the physical world. The datasets we create, the hardware we integrate, and the safety principles we establish for embodied systems will directly shape how these powerful foundation models see, understand, and interact with reality. This does raise a crucial question: in this race for high-impact, AGI-robust research, what happens to the slow, methodical, curiosity-driven science that has so often led to the biggest breakthroughs? It's a challenging but essential perspective for anyone building the eyes and brains of future machines. What's your take? How are you adapting your robotics research in light of these trends? You can read the full thread here: https://lnkd.in/e_J7mXMV #AGI #Robotics #ComputerVision #EmbodiedAI #FutureOfResearch #PhDLife #MachineLearning #AutonomousSystems