🤖 Should robots be generalists or specialists? At #IROS2026, @Ken_Goldberg & @andrea_bajcsy moderated one of the best debates in robotics. Matt Mason opened with a better question: “Should animals be humans?” 10 takeaways 🧵👇 1️⃣ @HarryXu12, for generalists: “Don’t build Excel. Build GPT.” We don’t know what we’ll need tomorrow, so buy generality just in case — and amortize it. The twist: GPT didn’t replace Excel. It learned to drive it. 2️⃣ @GeorgiaChal split “general” into 4 axes: Body. Scene. Contact. Task. “Breadth and reliability are measured per axis. One does not imply the other.” Her prediction: “The next task is the test.” 3️⃣ The most honest number of the day, from Georgia’s slides: A general model calling robot tools. Coarse block placement: 19/20 ✅ Precision insertion: 2/20 ❌ “Coarse pick and place works. Physical composition and precise contact not yet.” Tool use ≠ dexterity. 4️⃣ @KaterinaFragiadaki’s architecture: The generalist writes the curriculum. The specialist does the reps. Video → sim → RL on subgoals → specialist policies. No teleop. At deployment: “Fast specialists for the common case. Flexible generalists for the long tail.” 5️⃣ Specialist bench: “Fixture and conveyor are part of the intelligence.” A fixture is a prior. A conveyor is a scheduler. Strip them to look “general” and you’ve moved work onto a pricier learned controller. Optionality that hurts reliability isn’t intelligence. It’s cost. 6️⃣ A surgical robot doing one procedure 10,000 times learns exactly what can go wrong. “Generality is the enemy of certification.” From the floor: Are human drivers generalists or specialists? We certify generalists one task at a time. That’s what a license is. 7️⃣ Toshio Fukuda changed the register: “NON-AI matters.” Kizuki: help offered before it’s asked for. “Awareness comes from the body, not from the model.” Japan: ~29% aged 65+, ~570k care workers short by 2040. “If people live to 120, must our machines last that long?” 8️⃣ The floor fight: “A human is a generalist.” “We’re all human specialists.” The room accidentally re-derived pretrain-then-finetune. Veterans argued from what has shipped. Newcomers argued from what has scaled. Both are data. 9️⃣ The title asked for a side. The slides drew a stack: General planning + recovery on top. Fast, certifiable specialists below. A verifier is a fixture made of code. 🔟 The real question: How thick can the middle get before the body must specialize anyway? Generalist or specialist? Maybe the winning architecture is neither. It’s a stack. #IROS2026 #Robotics #PhysicalAI #RobotLearning #WorldModels

System 1 might be optional. Diffusion has no future. Humanoid control is 80% solved. 🌶️ 10 hot takes from the room at IROS 2026: 1️⃣ GPT-6 Astra is strong. It actually follows instructions, and that makes everything underneath it stronger. 2️⃣ Once a task is automated, it's just another function call. You may not need System 1 at all; call it only when needed. System 2 reasoning is already strong enough. 3️⃣ Some scenarios don't need a VLA or a world action model (WAM). Code-as-policy works. For dexterous hands: agent as a policy. 4️⃣ Guanya: LLMs like Astra are tools for doing robotics. Human imagination needs pretraining, and a pretrained LLM won't beat manual idea generation. 5️⃣ Design for zero-shot from day one. Zero-shot, start to finish. 6️⃣ Against in-context learning? Zero-shot from a foundation model right out of the gate? Even mapo tofu is fine-tuning. 7️⃣ Teleop can't do micro-motions like snapping a cookie. Vision sees no difference; the fingers make all the difference. Only RL gets you there, and probably not pure vision-based. 8️⃣ Humanoid control is 80% solved. The rest falls soon. 9️⃣ Workspace model, diffusion: no future. 🔟 The hand race: Sharpa wants to do it all. Wuji builds hands. Reasoning is moving up into frontier models, code and function calls. The unsolved part is moving down to the fingertips: contact, RL, beyond pure vision. Which take ages worst? 👇 #Robotics #EmbodiedAI #IROS2026 #HumanoidRobots #DexterousManipulation